You are on page 1of 20

JOURNAL OF EXPERIMENTAL SOCIAL PSYCHOLOGY 33, 451–470 (1997)

ARTICLE NO. JS971330

Categorization by Race: The Impact of Automatic


and Controlled Components of Racial Prejudice

Russell H. Fazio and Bridget C. Dunton

Indiana University

Received: May 28, 1996; revised: December 12, 1996; accepted: January 2, 1997

The present research examined the influence of both automatic and controlled processes
related to racial prejudice on the categorization of stimulus persons by race. Participants
judged the similarity of photos of individuals who varied in race, gender, and occupation.
These similarity ratings were subjected to a multidimensional scaling procedure. The
degree to which individuals weighted race in judging similarity was found to increase as a
function of the extent to which race was attitude-evoking for the individual and to decrease
as a function of motivation to control seemingly prejudiced reactions. It is suggested that
the attention of individuals for whom attitudes were automatically activated in response to
Black faces was automatically drawn to the race information, but that individuals
motivated to control prejudiced reactions actively resisted weighting race heavily. Exami-
nation of the latencies with which the similarity judgments were made provided support for
this account of the process underlying categorization by race. r 1997 Academic Press

The essential question underlying the present research concerns the categoriza-
tion of a stimulus person who can be thought of in multiple ways. Imagine
encountering a Black, female cashier at a grocery. Obviously, one could think of
this person as a Black, as a female, or as a cashier. Which of these possibilities
will predominate?
This issue must arise whenever a perceiver encounters any person—for a real
person is never a member of just one group but is always associated, at least to
some degree, with multiple social groups. Given that these various social groups
may themselves be associated with different traits, abilities, expectations, and

This research was supported by Research Scientist Development Award MH00452 and Grant
MH38832 from the National Institute of Mental Health to the first author. The authors thank Michael
Bailey, who developed and tested the software employed to present the high-resolution color images
as primes and to collect the response-latency data. We also thank Jennifer Bradley for so capably
serving as the experimenter for the second session and Edward Hirt and Eliot Smith for their helpful
comments on an earlier version of the manuscript.
Address correspondence and reprint requests to Russell H. Fazio, Dept. of Psychology, Indiana
University, Bloomington, IN 47401. E-mail: Fazio@indiana.edu.

451

0022-1031/97 $25.00
Copyright r 1997 by Academic Press
All rights of reproduction in any form reserved.

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
No. of Pages—20 First page no.—451 Last page no.—470
452 FAZIO AND DUNTON

evaluations, the outcome of the categorization process assumes considerable


importance. Our hypothetical target person may be construed quite differently if
she is viewed primarily in terms of her race, as opposed to her gender or
occupational role. What characteristics of the perceiver will allow one potential
categorization to predominate when other equally plausible categorizations are
available? In an effort to address this question, we examine the influence of both
automatic and controlled processes related to racial prejudice on the categoriza-
tion of such stimulus persons.
Attentional processes are known to exert a strong influence on categorization.
Within both cognitive (e.g., Medin & Schaffer, 1978; Nosofsky, 1986, 1988) and
more socially oriented models (Smith & Zarate, 1992), categorization is pre-
sumed to follow from the allocation of attention to particular dimensions of the
stimulus. Relatively more attention to a given stimulus attribute means that this
attribute will be relatively more influential in determining the perceived similarity
of the stimulus to exemplars of various categories. Hence, greater attention to an
attribute increases the likelihood that the stimulus person will be categorized
accordingly.
Obviously, attention to a given attribute can be fostered by contextual cues (see
Smith & Zarate, 1992, for a general discussion). The salience afforded a given
attribute as a result of its distinctiveness in the specific setting is one such example
(Mullen, 1991). For instance, the minority status of the lone Black or the lone
female in a group of White males is readily noted (e.g., Taylor, Fiske, Etcoff, &
Ruderman, 1978). Likewise, prior but recent activation of a potentially relevant
category confers an advantage to that category when a multiply categorizable
target person is encountered (e.g., Macrae, Bodenhausen, & Milne, 1995).
However, our focus in the present research is not on characteristics of the
situational context but on characteristics of the perceiver. More specifically, we
are concerned with the possibility that categorization is directly influenced by the
degree to which a given attribute is evaluatively laden for the perceiver.
Our reasoning regarding this possibility originates with research concerning the
influence of attitude accessibility on visual attention (Roskos-Ewoldsen & Fazio,
1992). This research demonstrated that objects toward which individuals hold
highly accessible attitudes automatically attract attention when they enter the
visual field. The work was based on a view of attitudes as object-evaluation
associations in memory (see Fazio, 1995, for a review). The associative strength
determines the accessibility of the attitude, i.e., the extent to which the evaluation
is capable of automatic activation from memory upon the individual’s encounter-
ing the object. In their series of experiments, Roskos-Ewoldsen and Fazio (1992)
examined the effects of differential attitude accessibility on visual attention.
Subjects were exposed to displays of six line drawings of objects. Objects toward
which the subjects possessed relatively accessible attitudes, which were termed
attitude-evoking objects, automatically and inescapably attracted attention. For
example, these attitude-evoking objects were: (a) more likely to have been
noticed after very brief exposures of each display (Experiments 1 and 2), (b) more

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
CATEGORIZATION BY RACE 453

likely to have been noticed incidentally during a task in which attending to these
items was neither required nor optimal (Experiment 3), and (c) more likely to
interfere with performance in a visual search task when presented as distractors
(Experiment 4).
On the basis of these findings, Roskos-Ewoldsen and Fazio (1992) suggested
that an attitude toward an object in the visual field may be activated from memory
at an early stage in the processing of visual information. Such activation then may
direct further attention to the visual stimulus so that, ultimately, the stimulus
receives sufficient processing for it to be consciously noticed and reported.
To continue with our example in the grocery setting, the categorizations of
‘‘female’’ and ‘‘Black’’ and ‘‘cashier,’’ as well as many other possibilities, would
be activated, to some extent, by the encounter with this target person. They may
not be activated equally. The stimulus characteristics themselves and the contex-
tual cues may make one or two possibilities predominate over others. However,
the various possible categorizations will receive some degree of activation. What
is important from the present perspective is that each of these possibilities may
itself be associated with an evaluation. Our argument is that the strength of any
such evaluative associations is critical. The implications of the Roskos-Ewoldsen
and Fazio (1992) findings are straightforward. The more attitude-evoking a
potential categorization, the more likely it is that attention will be drawn to that
particular attribute in the visual stimulus and, hence, the more likely that the
stimulus will be categorized accordingly.
The general hypothesis that attitude accessibility influences categorization has
received support in a series of experiments conducted recently by Smith, Fazio,
and Cejka (1996). These researchers drew an analogy between the presence of
multiple objects in the visual field (as in the displays employed by Roskos-
Ewoldsen & Fazio, 1992) and the existence of multiple potential categorizations
of an object in memory. The various categories associated with a stimulus object
may be partially activated when the object is encountered. Drawing on the visual
attention results, Smith et al. (1996) predicted that if the perceiver has a highly
accessible attitude attached to one of these categories, that category will attract
more attention and hence predominate over the alternatives. The experiments
focused on objects and people that could be thought of in multiple manners, i.e.,
ones that were multiply categorizable. For example, yogurt can be viewed as a
health food or as a dairy product, sunbathing as an activity that can lead to cancer
or as something one does at the beach, Pete Rose as a baseball player or as a
gambler. Using such triads consisting of a target and two alternative categoriza-
tions of the target, Smith et al. (1996) had research participants rehearse their
attitudes toward one of the two alternative categories and make animate versus
inanimate judgments regarding the other category. The participants later (in one
experiment, a full week later) were presented with the target words as cues to aid
their recall of these stimuli. The targets more effectively cued the categories for
which attitude accessibility had been enhanced via attitude rehearsal. For ex-
ample, when the accessibility of attitudes toward health food had been experimen-

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
454 FAZIO AND DUNTON

tally enhanced, yogurt was more likely to cue health food; when attitudes toward
dairy products had been rehearsed, yogurt was more likely to cue dairy product. In
an additional experiment, the time required to verify that a given target was a
member of a given category was similarly affected by the manipulation of attitude
accessibility. Participants were faster to verify category membership when the
accessibility of attitudes toward the category had been enhanced. Thus, converg-
ing evidence was obtained supporting the proposition that the potential category
toward which the individual had the more accessible attitude was more likely to
dominate the categorization process.
The present research tests this same general hypothesis, but in a very different
manner—with photos of people, instead of words, as the stimuli of interest. More
specifically, we sought to examine the hypothesis in the domain of racial attitudes.
Would participants’ racial attitudes affect their categorization of targets along a
race dimension? Both the findings regarding visual attention (Roskos-Ewoldsen
& Fazio, 1992) and the findings regarding the categorization of verbal stimuli
(Smith et al., 1996) point to the importance of the extent to which a possible
categorization is attitude-evoking or evaluatively laden. The more attitude-
evoking a potential categorization, the more likely it is that attention will be
drawn to that attribute. Greater attention to a given attribute, in turn, means that
the attribute will exert more influence as the stimulus is categorized. In the present
case, the more attitude-evoking race is for a given perceiver, the more likely it is
that the target’s race will automatically attract attention and that the perceiver will
categorize by race. This proposition constitutes the major hypothesis tested in the
current research.

AUTOMATIC AND CONTROLLED


COMPONENTS OF RACIAL PREJUDICE
The categorization process that we have outlined centers on the extent to which
race is attitude-evoking, i.e., the extent to which positive or negative evaluations
are automatically activated from memory upon exposure to a Black target person.
In order to assess this independent variable, we employed a procedure developed
and tested in recent research by Fazio, Jackson, Dunton, and Williams (1995) as
an unobtrusive measure of racial attitudes. The technique provides a measure of
the evaluations that are automatically activated from memory upon the presenta-
tion of faces of Black target individuals. Within this bona fide pipeline, as Fazio et
al. (1995) termed the procedure, the participant’s task on each trial is to indicate
the connotation of an evaluative adjective as quickly as possible. Does the
adjective mean ‘‘good’’ or ‘‘bad’’? The latency with which this judgment is made
and how it is affected by the prior presentation of a prime is the focus. Essentially,
the pattern of facilitation that is exhibited on positive versus negative adjectives
can provide an indication of the individual’s attitude toward the primed objects,
which in this case are high-resolution color images of Black and White undergradu-
ates. Relatively more facilitation on positive adjectives would be indicative of a
more positive attitude and relatively more facilitation on negative adjectives

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
CATEGORIZATION BY RACE 455

would be indicative of a more negative attitude. Obviously, what is of major


interest is the amount of facilitation on positive versus negative adjectives
when those adjectives are preceded by Black faces versus White faces. In fact, for
each individual participant, the effect size of this interaction between race of
photo and valence of adjective serves as an estimate of the individual’s racial
attitude.
A number of findings from this earlier research suggest that these unobtrusively
obtained estimates of attitude can be characterized as valid. First, Black and
White participants displayed very different patterns of facilitation and, as a result,
obtained very different scores. Just as one would expect, the White participants,
on the average, were characterized by more negative attitudes toward Blacks than
were the Black participants. Second, the considerable variability that was ob-
served among the White participants (some were characterized by relatively
extreme negative-attitude estimates, whereas others had scores similar to those
displayed by some of the Black participants) was meaningful. It was not simply
noise but instead proved predictive of other measures that were collected. The
unobtrusive estimates correlated significantly with a Black experimenter’s rating
of the quality of her interaction with each subject. Individuals identified
as having relatively negative attitudes toward Blacks according to the unobtrusive
priming measure were judged to have behaved more distantly. The measure also
correlated with participants’ judgments of the extent to which Blacks versus
Whites were more responsible for the escalation in tension and violence that
occurred in the city of Los Angeles following the announcement of the verdict of
‘‘not guilty’’ in the trial involving police officers accused of using excessive force
in arresting Rodney King. The more negative their attitudes toward Blacks, the
more the participants held Blacks relatively more responsible for the ensuing
riots.
All the participants in the present research underwent this priming procedure in
an initial experimental session. Thus, an estimate of the extent to which positivity
or negativity was automatically activated in response to Black faces was available
for each individual. Given that categorization by race also may be influenced by
more motivated, controlled processes, the research participants also completed
the ‘‘Motivation to Control Prejudiced Reactions Scale’’ (Dunton & Fazio, 1997).
This 17-item scale is an individual-difference measure of the extent to which
individuals are motivated to counter the influence of any seemingly prejudiced
reactions that they might experience. It includes such items as ‘‘I get angry with
myself when I have a thought or feeling that might be considered prejudiced,’’ ‘‘If
I have a prejudiced thought or feeling, I keep it to myself,’’ and ‘‘It’s never
acceptable to express one’s prejudices.’’All the scale items, as well as documenta-
tion of the scale’s reliability and predictive validity, are presented in Dunton and
Fazio (1997). Scores on the motivation scale have been found to moderate the
relation between attitude estimates derived from the bona fide pipeline procedure
and scores on the Modern Racism Scale (Fazio et al., 1995) and to moderate the
relation between such attitude estimates and direct, self-reported evaluations of

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
456 FAZIO AND DUNTON

Blacks (Dunton & Fazio, 1997). Individuals for whom negativity is automatically
activated are more likely to express such negativity verbally if their motivation to
control prejudiced reactions is relatively low.
To review, our conceptual framework suggests that the more attitude-evoking
race is for a given individual, the more likely it is that the individual will attend to
the target’s race and categorize by race. Our unobtrusive measure provides
precisely such an indication because it is based upon evaluations that are
automatically activated from memory. For some individuals, positivity is automati-
cally evoked in response to the Black faces; for others, negativity is automatically
evoked. Thus, we predict a curvilinear relation between the attitude estimates
derived from the priming procedure and categorization by race. Those individuals
for whom race is attitude-evoking should have their attention automatically drawn
to the target’s skin color. However, some individuals—those for whom motivation
to control prejudiced reactions is relatively high—may not appreciate their
attention being drawn to race in this way and may actively seek other means of
categorizing the target person.
In order to assess categorization by race, we employed a similarity-rating
procedure and multidimensional scaling—a technique that has been used fre-
quently to study attention and categorization (see Nosofsky, 1992). In a second,
ostensibly unrelated session 1 week after the assessment of racial attitudes, the
research participants were presented with pairs of photos and asked to rate the
similarity of the individuals in each pair. All the photos contained contextual cues
such as uniforms, tools, furniture, or background setting that served to identify the
individual’s occupation. Both males and females and Blacks and Whites were
presented. So, the photos varied in race, gender, and occupation and, hence, could
be categorized in a number of ways. We presumed that race would emerge
as one of the dimensions that served as a basis for judging similarity but that
individuals would vary in the extent to which they weighted race when judging
similarity. Our primary concern is with the extent to which this variability in the
use of the race dimension can be predicted from our unobtrusive estimates of
automatically activated racial attitudes and from any motivation to control
prejudiced reactions.
METHOD
Subjects
A total of 60 Indiana University undergraduates participated in the initial session of the experiment
either for partial fulfillment of an introductory psychology course requirement or for monetary
compensation. Fifty-seven returned 1 week later for the second experimental session. In addition, the
data from three subjects were eliminated, resulting in a final sample size of 54. Two were eliminated
because they committed excessive errors on the adjective-connotation tasks performed during the
initial session. The third was Black and was eliminated for this reason.

Stimulus Materials
Session 1. Forty-eight color photographs of White, Black, Asian, and Hispanic male and female
undergraduates served as primes. The photographs were digitized as 256-color, 640 3 480 resolution
images for presentation on the computer screen. All photographs were head shots taken against a

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
CATEGORIZATION BY RACE 457

common background. These undergraduates had been paid five dollars to have their pictures taken and
signed consent forms permitting the use of their photographs in the laboratory’s research.
Session 2. Twenty-four color photographs of individuals in various occupations served as stimuli for
the paired similarity judgments. The individuals in the photographs were intended to vary on at least
three dimensions: race (12 White and 12 Black), gender (12 male and 12 female), and occupational
status. Included were such occupations as police officer, bricklayer, carpenter, mail carrier, cashier,
professor, pharmacist, and business person. Again, the photographs were presented as 256 color,
640 3 480 resolution images.

Procedure
Session 1. The procedure closely followed that employed by Fazio et al. (1995). Participants were
told that the experiment involved the judgment of word meaning as an automatic skill and that they
would be performing a variety of tasks through the course of the experimental session. The procedure
consisted of five phases, with the fourth phase being the priming task of interest.
The first phase involved an adjective-connotation task and was intended to provide baseline
response-latencies for the adjectives that would later be used as target words in the priming phase of
the experiment. Participants were presented with a series of words on the computer screen and
instructed to press a key labeled ‘‘good’’ if the word had a positive meaning or to press a key labeled
‘‘bad’’ if the word had a negative meaning. They were asked to maximize the speed and accuracy with
which they made these word-meaning judgments. Each of the 24 adjectives—12 positive (e.g.,
appealing, attractive, likable, pleasant) and 12 negative (e.g., annoying, awful, offensive, repulsive)
was presented in a randomized order to each subject. A row of asterisks preceded each adjective and
served as a warning signal that a word was about to appear on the screen. Each adjective remained on
the screen for a maximum of 1.75 s, but disappeared as soon as the subject responded. A 2.5-s interval
separated each trial. For each adjective, the response was recorded as well as the latency with which
the response was made. Participants first performed a practice block consisting of 12 adjectives to
familiarize themselves with the procedure and to confirm that they understood the instructions. They
then participated in two blocks of actual trials, each block consisting of all 24 adjectives. For each
individual, the average latency across the two trials for a given adjective served as the baseline latency
for that adjective.
The next two phases of the experiment were intended to prepare the participants for the upcoming
priming task, as well as to bolster the cover story that the experiment involved the judgment of word
meaning as an automatic skill. Participants were informed that they would be performing a
face-learning task followed by a face-detection task.
The second (face-learning) phase involved the ability to learn faces for a later face-recognition
phase. Participants were instructed to attend to a series of faces as they appeared on the computer
screen. They were told to remember the faces in order to demonstrate their recall of the faces during a
later recognition task. The stimuli consisted of 16 high-resolution color photographs of White, Black,
Asian, and Hispanic male and female undergraduates. Each photograph was presented twice (once
during each of two blocks).
Recognition of the faces was tested in the third (face-detection) phase. The 16 photographs from the
previous task, as well as 16 new (not previously presented) photographs, were presented. The
participants were instructed to press a key labeled ‘‘yes’’ when they recognized the face as being from
the previous task or to press a key labeled ‘‘no’’ when they did not recognize the face as being from the
previous task.
The fourth phase of the experiment involved the actual priming task and provided the target
latencies of interest. The participants were informed that the previous tasks would now be combined.
As part of the cover story, they were told that, if judging the connotation of adjectives is truly an
automatic skill, individuals should be able to perform just as well as in the very first phase of the
experiment even if they had to do something else at the same time—the something else in this case
being to learn faces. Thus, subjects were led to believe that this phase of the experiment involved both
the learning of the faces and the judgment of the connotation of the adjectives. The procedure was

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
458 FAZIO AND DUNTON

identical to the first phase of the experiment except that the row of asterisks was replaced with a
photograph of a face. The instructions were also identical except that subjects were told they would be
tested for their recall of the faces presented during this phase and should therefore carefully attend to
the faces. On each trial, the prime (photograph) was presented for 315 ms, followed by a 135-ms
interval before presentation of the target adjective. Again, a maximum of 1.75 s to respond was
allowed and trials were separated by a 2.5 s interval. A total of four blocks was presented. Each of the
48 primes was presented during each of the four blocks. Each prime was followed by two positive and
two negative adjectives. The critical primes were 16 Black and 16 White faces which had been paired
with one another, such that the paired faces were followed by the same four adjectives. The remaining
16 primes consisted of matched pairs of White and Asian or Hispanic faces. These filler trials were
included to reduce the overall proportion of Black faces that was presented.
The fifth phase of the procedure involved a recognition task to further bolster the cover story.
Participants were presented with the 48 faces presented during the priming phase along with 48 filler
faces not presented during that phase. Again, as in the third phase, participants were instructed to
identify the faces they recalled from the previous priming task. The procedure and instructions were
identical to those in the third phase.
Session 2. Participants returned one week later for an ostensibly unrelated second session conducted
by a different experimenter. They were informed that they would be presented with pairs of
photographs of individuals in various occupations on the computer screen. Their task was to rate the
similarity of the individuals in each pair on a scale from 1 to 9 with 1 being ‘‘not at all similar’’ and 9
being ‘‘extremely similar.’’ Each of the 24 stimulus photographs was paired once with each of the
other 23 stimulus photographs for a total of 276 paired similarity judgments. Each pair of photographs
remained on the screen for a maximum of 5 s, but disappeared as soon as the participant responded.
For each trial, the similarity judgment was recorded, as well as the latency with which the judgment
was made. The 276 trials were divided into six blocks of 46. Participants were encouraged to rest
between blocks.
Following the paired similarity procedure, participants completed the ‘‘Motivation to Control
Prejudiced Reactions Scale’’ (Dunton & Fazio, 1997). They were then debriefed and dismissed.

RESULTS
Attitude Estimates
Attitude estimates were computed in the same manner employed by Fazio et al.
(1995). For each subject, baseline latencies for each adjective were computed by
averaging the latencies for the two presentations of that adjective during the first
phase of session one. The baseline latency for each adjective was then subtracted
from the latency for each presentation of that target adjective during the priming
task to arrive at a facilitation score for each face/adjective combination. Mean
facilitation scores on the two positive adjectives and on the two negative
adjectives then were computed for each face. Only correct responses to target
adjectives (average percent correct was 97%) were included in the computations.
Even with this data reduction, multiple observations remain for each subject—
that is, mean facilitation scores for 16 White faces and 16 Black faces. Thus, it is
possible to examine the interaction of race of photo 3 valence of adjective for
each and every individual participant. Following Fazio et al. (1995), the effect
size of this interaction served as our estimate of the individual’s attitude. More
negative scores reflect a pattern of facilitation indicating greater negativity toward

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
CATEGORIZATION BY RACE 459

TABLE 1
CORRELATIONS BETWEEN DIMENSION COORDINATES AND VARIOUS ATTRIBUTES OF THE STIMULI

Dimension

Attribute 1 2 3 4

Status .89 2.24 .15 2.21


Independence .07 .71 .23 .18
Race 2.19 .13 .96 2.05
Gender 2.11 .02 .08 .89

Blacks.1 The attitude estimates ranged from 2.89 to .54, with a mean of 2.02 and
a SD of .29.

Multidimensional Scaling
The similarity ratings were subjected to a multidimensional scaling procedure,
specifically INDSCAL. A four-dimensional solution emerged, characterized by a
multiple R 2 of .621 and a stress value of .163. Dimensions one through four
accounted for 24.48, 17.31, 10.48, and 9.83% of the variance, respectively.
In order to facilitate the interpretation of the multidimensional space, an
independent sample of 34 judges rated each of the 24 stimulus photos along
various scales. The judges rated the status of the occupation depicted in each
photo. They also rated the independence of each occupation, i.e., the extent to
which it involved working alone or with others. Both of these ratings proved to
relate to one of the dimensions. Table 1 displays a correlation matrix involving the
correlation between a given attribute of the photos and the coordinates of the
photo on a given dimension. As is evident from the correlations, the first
dimension clearly refers to occupational status. Dimension 2 correlated highly

1 Use of the race of photo 3 valence of adjective interaction as the basis for the computation of the

attitude estimate has the effect of defining racial attitudes in a relative sense, i.e., as the nature of the
evaluations activated in response to Black faces relative to the evaluations activated in response to
White faces. We believe this operationalization to be defensible on both conceptual and methodologi-
cal grounds. Conceptually, the estimation procedure captures the essence of racial prejudice, which is
that Blacks are responded to or treated differently from Whites. Methodologically, the inclusion of the
data from trials involving White primes is necessary as a means of controlling for the influence of
presenting any face on the activation of positive versus negative evaluations. Moreover, any estimate
of attitude based solely upon the degree to which Black faces facilitated responding to positive versus
negative adjectives fails to correct for the extent to which the set of positive and negative adjectives
might differ in their potential for facilitation. In fact, our experiments have consistently revealed a
main effect of adjective valence (see Fazio et al., 1995). Thus, focusing on the race 3 adjective
interaction has the advantage of controlling for individual differences in this adjective valence main
effect. In any case, our empirical work clearly demonstrates the value of the interaction-based
estimates of racial attitudes. When the analyses reported in the present paper were conducted with the
effect size of the adjective valence difference for Black prime trials alone as the estimate of racial
attitudes, none of the reported effects was obtained.

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
460 FAZIO AND DUNTON

FIG. 1. Frequency distribution of weight given to the race dimension. The squared weight
indicates the proportion of variance in a participant’s similarity matrix that is accounted for by the race
dimension.

with our judges’ ratings of independence. Dimension 4 correlated with gender.


And most important, dimension 3 clearly refers to race.
One advantage of INDSCAL is that it yields not only this multidimensional
space but also a set of weights for each individual respondent showing the extent
to which the individual weighted each dimension in making his or her similarity
ratings. When squared, these weights are readily interpretable; they indicate the
proportion of variance in the given respondent’s similarity ratings that is ac-
counted for by each dimension. A frequency distribution of the squared subject
weights for the race dimension is displayed in Fig. 1.

Predicting Use of the Race Dimension


As is evident from Fig. 1, considerable variability exists across the respondents
in the extent to which they use the race dimension when judging similarity. Some
subjects use it little; others use it much more. The question is whether this
variability relates to the variability that was witnessed with respect to our measure
of automatically activated racial attitudes. We conducted a multiple regression
predicting the squared subject weights from our attitude estimates and the squared
values of these attitude estimates. Although the linear attitude term did not relate
to the subject weights for the race dimension, R 2 5 .006, F , 1, the quadratic term
produced a significant increase in the proportion of variance explained, change in
R 2 5 .095, F(1, 51) 5 5.36, p , .025. When scores on the Motivation to Control

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
CATEGORIZATION BY RACE 461

FIG. 2. Regression curves predicting weight assigned to the race dimension as a function of the
unobtrusive attitude estimates and motivation to control prejudiced reactions.

Prejudiced Reactions Scale were added to the equation, the multiple correlation
again increased significantly, change in R 2 5 .102, F(1, 50) 5 6.41, p , .02.2
Cross-product terms reflecting the interaction between attitude and motivation did
not increase the multiple correlation significantly when subsequently entered into
the equation.
The curvilinear regression lines relating the attitude estimates to the subject
weights for the race dimension are displayed in Fig. 2. The two curves correspond
to values 1 SD above and below the mean on the Motivation to Control Prejudiced
Reactions Scale. The observed relation matches the prediction well. Those
individuals for whom negativity was automatically activated in response to
Blacks have weights indicating that they used the race dimension heavily in
making their similarity ratings. The same is true of those for whom positive
attitudes are automatically activated. On the other hand, those participants who
experienced little or no automatic attitude-activation in response to the Black

2 Analyses are reported in terms of squared subject weights, which have the advantage of being

interpretable as the proportion of variance in the similarity ratings that is accounted for by the given
dimension. However, analyses of the raw weights revealed nearly identical results: for the quadratic
term, change in R 2 5 .067, F(1, 51) 5 3.69, p 5 .06; for the motivation variable, change in R 2 5 .095,
F(1, 50) 5 5.66, p , .025.

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
462 FAZIO AND DUNTON

faces presented in the first session weighted race less when judging similarity in
the second session. The nature of the observed main effect of the controlled
component also is readily apparent from Fig. 2. Those participants who described
themselves as being highly motivated to control prejudiced reactions did not use
the race dimension as heavily as did those who reported relatively low motivation.
Analyses examining the subject weights for the other three dimensions were
also conducted. The attitude estimates did not relate to the use of these dimen-
sions. However, scores on the motivation scale did relate to the occupation
dimensions. The correlation between motivation scores and the sum of the
squared subject weights for the two dimensions related to occupation was .29, p ,
.05. This finding suggests that the more motivated individuals were to control
seemingly prejudiced reactions, the more likely they were to shift their attention
from race to a consideration of the occupational cues.
An additional regression analysis yielded evidence consistent with the implica-
tions of this latter finding. Weights for the race dimension were predicted from the
independent variables after controlling for the subject weights associated with
each of the other three dimensions. These additional subject weights strongly
related to the weights for the race dimension, R 2 5 .377, F(3, 50) 5 10.10, p ,
.001; greater use of the race dimension was associated with less use of the two
dimensions concerning occupation and with more use of the gender dimension.
Thus, inclusion of the other dimensions in the equation served to reduce the error
variance markedly as well as controlling for their influence. When the use of the
other dimensions was controlled in this way, the curvilinear relation between
the attitude estimates and use of the race dimension remained apparent. In fact,
the quadratic term was highly significant, t(47) 5 3.90, p , .001. However, the
motivation variable no longer contributed to the prediction of the subject weights
for the race dimension, t , 1. Thus, motivation to control prejudiced reactions
appears to have influenced use of the race dimension via a corresponding shift in
attention to the occupational cues.
Let us consider the findings in light of the attentional mechanism that was
delineated earlier. According to our framework, the data suggest that upon
presentation of a Black target, attitudes toward Blacks were automatically
activated for some participants. As a result, their attention was automatically
drawn to the race information. Greater attention to the race information led these
individuals to weigh race more heavily in judging similarity. However, some such
individuals apparently did not appreciate their attention having been drawn to
race in this way. Those motivated to control prejudiced reactions seem to have
exerted some effort to counter this influence and, ultimately, used the race
dimension less.

Judgment Latencies
With the above explanation for the findings in mind, we undertook additional
analyses of the available data in the hope of uncovering some corroborating

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
CATEGORIZATION BY RACE 463

evidence for this particular account of the underlying process. Unbeknownst to


the participant, the computer that was displaying the stimuli and recording the
similarity ratings also was measuring the latency of those responses. The analyses
that proved most informative involved latencies on trials in which at least one of
the two targets was Black. These trials proved very interesting.
Given the length of the similarity task, 276 trials, we were concerned that
strategies, efficiencies, and patterns of response latency might change over the
course of the task. So, we conducted separate analyses on the first third of the
trials and the last third of the trials. In these regression analyses, we treated the
average latency on the trials involving one or two Black targets as the dependent
variable. (This average was computed after first conducting a reciprocal transfor-
mation of the latencies so as to reduce the influence of the skewness typically
observed with such latency data.) The first predictor that we entered into the
equation was the mean transformed latency on trials involving two White targets;
this variable provided a means of controlling for individual differences in baseline
speed of responding. So, the effects of interest concern latency on trials involving
a Black target relative to latencies on trials involving two Whites. The next
variables entered into the equation were our attitude estimate, its squared value
(i.e., the quadratic term), and the motivation-scale score. On the first third of the
trials, the analysis revealed a significant quadratic effect of the racial attitude
estimates, t(49) 5 2.35, p , .025, as well as an effect of motivation, t(49) 5 2.01,
p , .05. Crossproduct terms involving the interaction between attitude and
motivation did not increase the multiple correlation when subsequently entered
into the equation, t’s , 1.3
The regression curves are depicted in Fig. 3. The shape of the curves indicates
that those individuals for whom race was attitude-evoking were faster to respond
than those for whom race was more neutral. Apparently, their attention was
automatically drawn to the race information, allowing them to respond relatively
quickly. Moreover, we see that those participants motivated to control prejudiced
reactions exerted some time-consuming effort in responding, presumably to
counter the influence of attention to race. The more motivated took relatively
longer to respond, suggesting that the motivation prompted a controlled process.

3 The effects that were observed appear to have been driven largely by the trials involving

mixed-race targets. When separate regression analyses were conducted on the mixed-race trials and
the trials involving two Black targets, the former yielded a statistically significant quadratic
effect of the attitudes estimates, t(49) 5 3.011, p , .005, as well as a significant effect of motivation,
t(49) 5 2.42, p , .02, whereas the latter did not, t’s , 1. Whether the different outcomes reflect a
meaningful difference in the manner in which participants approached these two kinds of trials cannot
be discerned with any confidence. The two kinds of trials also differed in their frequency. Across the
entire task of 276 trials, during which each photo is paired with every other photo, 144 involve
mixed-race targets and 66 involve two Black targets. Thus, for the first third of the trials, the number of
observations available for Black–Black targets is only 22, compared to 48 for Black-White targets,
which may have contributed to the less reliable findings observed for the trials involving two Black
targets.

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
464 FAZIO AND DUNTON

FIG. 3. Regression curves predicting mean latency (from the first third of the trials) to respond on
trials involving one or more Black targets as a function of the unobtrusive attitude estimates and
motivation to control prejudiced reactions.

By the last third of the trials, these relations no longer held true. Neither the
curvilinear main effect of attitude nor the main effect of motivation was signifi-
cant. However, in our examination of these data, it became obvious that the
respondents’ speed of responding had improved markedly over the course of the
task. The average latency for trials involving at least one Black target was 2143
ms on the first third of the trials; on the last third of the trials, the mean latency
improved to 1625 ms, t(53) 5 9.23, p , .001. Latencies on the trials involving
two White targets displayed similar improvement, M’s of 2195 and 1692 ms,
t(53) 5 8.90, p , .001.
As a result of this observation, we became interested in the possibility that the
variables of interest might relate to the extent to which respondents’ latencies
improved from the first third to the last third of the trials. A regression analysis on
these gain scores revealed an interesting pattern. The gain scores for the trials
involving Blacks were predicted from the gain scores for the trials involving two
Whites, the attitude estimates, the squared value of the attitude estimates,
motivation scores, and finally interactions between the attitude estimates and
motivation. Neither the curvilinear main effect of attitude nor the main effect of
motivation was significant. However, a statistically significant interaction emerged

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
CATEGORIZATION BY RACE 465

FIG. 4. Regression curves predicting the mean gain (from the first third to the last third of the
trials) in latency to respond on trials involving one or more Black targets as a function of the
unobtrusive attitude estimates and motivation to control prejudiced reactions.

between the quadratic attitude variable and motivation, t(47) 5 2.05, p , .05.4
The form of this interaction can be discerned from Fig. 4, which displays
regression curves for motivation scores 1 SD above and 1 SD below the mean
scale score. Respondents characterized by relatively low motivation improved,
regardless of their attitude estimates. That is, regardless of whether race was or
was not attitude-evoking for them, these individuals displayed latency gains. The
high-motivation, neutral-attitude participants also displayed this level of improve-
ment. However, the high-motivation individuals for whom race was attitude-
evoking displayed relatively little improvement. Although their relatively slow
latencies to respond on the first third of the trials suggest that these more
motivated people actively resisted the influence of having had their attention
automatically drawn to race, the race information apparently continued to attract
their attention more than they wished. As a result, they were less apt than other

4 Additional analyses indicated that both the trials involving mixed-race targets and those involving

two Black targets contributed to this effect. When separate analyses were conducted, the interaction
term reached a marginal level of statistical significance for both types of trials, t(47) 5 1.77, p , .09
and t(47) 5 1.84, p , .08, respectively.

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
466 FAZIO AND DUNTON

respondents to display any substantial improvement in the speed with which they
could judge the Black targets.
These two latency findings—the pattern on the first third of the trials and the
differential gains in speed of responding that people made—provide at least
tentative corroboration for the process that we believe to be underlying the extent
to which individuals used race when judging similarity. For people for whom race
is attitude-evoking, attention is automatically drawn to the race information,
permitting them to judge similarity relatively quickly. Those for whom motivation
to control prejudiced reactions is high work to counter this automatic attentional
effect; they appear to search for other bases for judging similarity. But, such effort
requires additional time on their parts. Moreover, the automatic attraction of
attention to race that these people experience and their ensuing efforts to consider
attributes other than race appear to prevent these individuals from improving
much in the speed with which they perform the judgmental task.

DISCUSSION
The results indicate that characteristics of the perceiver play an important role
in how stimuli are categorized. The multidimensional scaling of participants’
similarity judgments revealed considerable variability in the extent to which
people relied upon race, gender, and/or occupational cues when judging the
similarity of the stimulus persons. Although some people used race extensively as
a basis for judging similarity, others did not. Consistent with our hypotheses, this
variability in the use of race was predictable from characteristics of the perceiver.
Individuals for whom race was attitude-evoking weighted race more heavily in
judging similarity. Individuals motivated to control seemingly prejudiced reac-
tions, on the other hand, used race relatively less.
Thus, categorization by race was influenced by both automatic and controlled
processes. The automatic process appears to have involved the automatic activa-
tion of racial attitudes upon presentation of a Black target person—activation that
had the consequence of drawing further attention to the race of the target. The
underlying process seems to center on a cyclical, reverberating pattern of
activation involving the evaluation and the attribute with which it is associated
(race in this case). Some minimal activation of ‘‘Black’’ upon presentation of the
target evokes the associated attitude, which then leads to further activation of the
attribute. (See Smith et al., 1996, for discussion of a connectionist model of
accessible attitudes that involves such bidirectional, recurrent activation.) Having
attention automatically drawn to race in this way provided a quick and easy means
of judging similarity. Participants for whom race was attitude-evoking were able
to respond relatively quickly but, as a consequence, their similarity judgments
were influenced relatively more heavily by race than by equally reasonable
consideration of occupation or gender.
The controlled process involved in categorization by race appears to stem from
a motivation to avoid seemingly prejudiced reactions. Individuals who described
themselves as so motivated weighted race less but doing so was associated with a

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
CATEGORIZATION BY RACE 467

cost. They required more time to make their judgments. Apparently, their
reluctance to base a judgment on race prompted them to search for and consider
other attributes of the stimulus persons.
This ‘‘search for alternatives’’ appears to have been especially problematic for
those individuals who reported relatively high motivation to control prejudiced
reactions yet had their attention automatically drawn to the race information.
Motivated people for whom race was attitude-evoking were the one subset of
participants who failed to display a substantial gain in the speed with which they
made their judgments over the course of the lengthy task. The relatively less
motivated became more rapid at making the similarity judgments with repeated
presentations of the stimuli. So did the highly motivated people for whom race
was not attitude-evoking. But those motivated individuals whose attention was
drawn to race by virtue of race being attitude-evoking for them were unable to
develop such efficiency. They were continually faced with the task of directing
their attention away from race.
Our explanation for the findings centers on the race of Black target persons
having automatically attracted attention for some participants. The latency data,
in particular, support this notion. Relative to the latency with which they judged
stimulus pairs involving two White targets, participants for whom race was
attitude-evoking were faster to judge stimulus pairs that included Black targets.
Although we believe attention to have been a function of a particular value of the
race dimension (i.e., Black targets as opposed to White targets), it is important to
recognize that, over the course of the lengthy similarity task, such a force would
develop into greater attention to the stimulus dimension of race. Each White target
eventually was paired with each of 12 Black targets. On any such trial occurring
early in the task, the race of the Black target would have attracted attention and
encouraged categorization of the target as Black, which is likely to have had the
consequence of encouraging categorization of the specific White target in terms of
race as well. Having thus considered this White target in terms of race, partici-
pants are likely to have viewed later presentations of the same target in the same
way, even when the target was paired with another White target person. Thus, the
repeated presentations of targets during the course of the task is likely to have
contributed to the emergence of the greater use of the race dimension that was
evident in the multidimensional scaling results. However, our theoretical frame-
work and the latency data lead us to believe that the process began with attention
being automatically drawn to the race of the Black targets.
The present research findings illustrate the importance of attitude accessibility
for attention and categorization. When they encounter a target person who is
categorizable in multiple ways, people do not necessarily ‘‘see’’ the same person.
Who they see is at least somewhat dependent on their own characteristics. The
notion that perception is a function of the perceiver is by no means novel; this
principle formed the essence of the New Look movement (Bruner, 1957).
However, the present findings demonstrate how accessible attitudes, in particular,
can influence the way in which other people are viewed. Like the findings of

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
468 FAZIO AND DUNTON

Roskos-Ewoldsen and Fazio (1992) and Smith et al. (1996), our research
illustrates the power of attitudes that are highly accessible from memory, i.e.,
capable of automatic attitude activation. Just as Fazio, Roskos-Ewoldsen, and
Powell (1994) recently argued, the influence of accessible attitudes on informa-
tion is not limited to higher order cognitive processes that involve the interpreta-
tion, judgment, and integration of attribute information. Such attitudes also have
the potential to influence more basic processes of perception, attention, and
categorization.

Implications for Behavior


Both people for whom positivity was automatically activated in response to
Black faces and those for whom negativity was activated displayed a bias toward
categorizing the targets by race. What does this imply for an actual interaction
with a Black target? Someone characterized by automatically activated positivity
is very likely to categorize the target as Black, but the outcome should be
relatively benign, because judgment and behavior will be influenced by a positive
attitude. Thus, the bias in such cases is without accompanying prejudice. How-
ever, the same cannot be said about an individual for whom negativity is
automatically activated. Such an individual is likely to categorize the target as
Black, as opposed to female, cashier, or whatever else. In fact, having done so
probably activates the perceivers’ negative attitudes toward Blacks even more.
This automatically activated negativity will then influence perceptions of this
target person in this immediate situation and, ultimately, judgments about and
behavior toward the target. Thus, these cases involve bias with prejudice.
However, just as Devine (1989) has postulated, some individuals for whom
negativity is automatically activated may be motivated to counter the effects of
that negativity. Some people seem to carefully and deliberately monitor their
verbal expressions and behavior so as to avoid a prejudiced response. As noted
earlier, evidence supporting such a moderating role of motivational concerns has
been observed with respect to the relations between automatically activated
attitudes and both scores on the Modern Racism Scale (Fazio et al., 1995;
Experiment 4) and self-reported evaluations of Blacks (Dunton & Fazio, 1997). A
parallel influence of motivation to control seemingly prejudiced reactions was
observed in the present research with respect to some people’s reluctance to weigh
race heavily when judging the similarity of two stimulus persons.
The influence of automatic and controlled components of racial prejudice can
be considered from the perspective of the MODE model (Fazio, 1990), which
posits that Motivation and Opportunity serve as DEterminants of the extent to
which judgments and behavior follow from a relatively spontaneous process
involving the influence of automatically activated attitudes or a more deliberative
process involving effortful analysis. According to the model, both motivation and
opportunity are required for more deliberative analysis. In the case of race-related
judgments and behavior, the relevant motivation involves a desire to control
seemingly prejudiced reactions. Whether any attempt to control race-related

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
CATEGORIZATION BY RACE 469

expressions and behavior is successful will depend upon the opportunity factor.
Some behaviors are more easily monitored and controlled than others. Respond-
ing to the Modern Racism Scale, verbally expressing feelings toward Blacks, and
providing ratings of the similarity between two stimulus persons are examples of
activities that one can control. Judgments and behaviors that provide the opportu-
nity for controlling one’s response should be predictable from the joint influence
of automatically activated attitudes and motivation to control prejudice. However,
as opportunity decreases, because the behavior is not easily controllable, the
behavior should be less influenced by motivational concerns and more directly
and singly influenced by any automatically activated evaluations. Nonverbal
channels are an excellent example of behaviors that are difficult to control.
Consistent with the implications of the MODE model, strong relations have been
observed between automatically activated attitudes toward Blacks and such
nonverbal behaviors as eye contact during a subsequent interaction with a Black
target person (Dovidio, Brigham, Johnson, & Gaertner, 1996).
In sum, the MODE model suggests that race-related judgments and behavior
will depend upon automatically activated attitudes, any motivation that the
individual may experience to behave in a nonprejudiced manner, and finally the
extent to which the behavior in question and the situation provide an opportunity
for such a motivated process to succeed.

REFERENCES
Bruner, J. S. (1957). On perceptual readiness. Psychological Review, 64, 123–152.
Devine, P. G. (1989). Stereotypes and prejudice: Their automatic and controlled components. Journal
of Personality and Social Psychology, 56, 5–18.
Dovidio, J. F., Brigham, J. C., Johnson, B. T., & Gaertner, S. L. (1996). Stereotyping, prejudice, and
discrimination: Another look. In N. Macrae, M. Hewstone, & C. Stangor (Eds.), Foundations of
stereotypes and stereotyping. New York: Guilford.
Dunton, B. C., & Fazio, R. H. (1997). An individual difference measure of motivation to control
prejudiced reactions. Personality and Social Psychology Bulletin, 23, 316–326.
Fazio, R. H. (1990). Multiple processes by which attitudes guide behavior: The MODE model as an
integrative framework. In M. P. Zanna (Ed.), Advances in experimental social psychology. New
York: Academic Press.
Fazio, R. H. (1995). Attitudes as object-evaluation associations: Determinants, consequences, and
correlates of attitude accessibility. In R. E. Petty & J. A. Krosnick (Eds.), Attitude strength:
Antecedents and consequences (pp. 247–282). Hillsdale, NJ: Erlbaum.
Fazio, R. H., Jackson, J. R., Dunton, B. C., & Williams, C. J. (1995). Variability in automatic
activation as an unobtrusive measure of racial attitudes: A bona fide pipeline? Journal of
Personality and Social Psychology, 69, 1013–1027.
Fazio, R. H., Roskos-Ewoldsen, D. R., & Powell, M. C. (1994). Attitudes, perception, and attention. In
P. M. Niedenthal & S. Kitayama (Eds.), The heart’s eye: Emotional influences in perception and
attention (pp. 197–216). New York: Academic Press.
Macrae, C. N., Bodenhausen, G. V., & Milne, A. B. (1995). The dissection of selection in person
perception: Inhibitory processes in social stereotyping. Journal of Personality and Social
Psychology, 69, 397–407.
Medin, D. L., & Schaffer, M. M. (1978). Context theory of classification learning. Psychological
Review, 85, 207–238.

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb
470 FAZIO AND DUNTON

Mullen, B. (1991). Group composition, salience, and cognitive representations: The phenomenology
of being in a group. Journal of Experimental Social Psychology, 27, 297–323.
Nosofsky, R. M. (1986). Attention, similarity, and the identification-categorization relationship.
Journal of Experimental Psychology: General, 115, 39–57.
Nosofsky, R. M. (1988). Similarity, frequency, and category representations. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 14, 54–65.
Nosofsky, R. M. (1992). Similarity scaling and cognitive process models. Annual Review of
Psychology, 43, 25–53.
Roskos-Ewoldsen, D. R., & Fazio, R. H. (1992). On the orienting value of attitudes: Attitude
accessibility as a determinant of an object’s attraction of visual attention. Journal of Personality
and Social Psychology, 63, 198–211.
Smith, E. R., Fazio, R. H., & Cejka, M. A. (1996). Accessible attitudes influence categorization of
multiply categorizable objects. Journal of Personality and Social Psychology, 71, 888–898.
Smith, E. R., & Zarate, M. A. (1992). Exemplar-based model of social judgment. Psychological
Review, 99, 3–21.
Taylor, S. E., Fiske, S. T., Etcoff, N. L., & Ruderman, A. J. (1978). Categorical bases of person
memory and stereotyping. Journal of Personality and Social Psychology, 36, 778–793.

JESP 1330
@xyserv1/disk3/CLS_jrnl/GRP_jesp/JOB_jesp97ps/DIV_319z02 debb

You might also like