Applied Animal Behaviour Science 151 (2014) 84–93

Contents lists available at ScienceDirect

Applied Animal Behaviour Science

journal homepage: www.elsevier.com/locate/applanim

Explosives detection by military working dogs: Olfactory

generalization from components to mixtures
Lucia Lazarowski ∗ , David C. Dorman
Department of Molecular and Biomedical Sciences, North Carolina State University, College of Veterinary Medicine, Raleigh, NC, USA

a r t i c l e i n f o a b s t r a c t

Article history: The training of scent detection dogs using samples of explosives or their chemical precursors
Accepted 22 November 2013 is a well-established and documented practice. However an area of canine odor detection
Available online 1 December 2013
that remains under-studied regards a trained dog’s perception of an explosive odor when
more than one odorant is combined to produce a mixture. The first objective of our study
Keywords:
was to determine whether training adult Labrador Retrievers (n = 20) to detect the scent
Dog
of chemically pure potassium chlorate (PC) was sufficient to produce generalization to PC-
Olfactory
Explosives based explosive mixtures that contained a novel component. We found that the majority
Odor mixtures of dogs (87%) trained with pure PC alone did not correctly signal the presence of one or
more of four PC-based explosive mixtures. Our second objective was to determine whether
training dogs using the separated components found in the PC-based explosives would
subsequently enhance detection. Dogs were then trained using a novel static odor delivery
device that safely segregated the PC and non-PC components and presented a merged odor
to the dog. A statistically significant improvement in percentage of dogs detecting PC-based
mixtures after training with the separated components compared to training with PC alone
was seen with Mixture 1 (27–100%, P < 0.0001), Mixture 2 (40–81%, P = 0.0229), Mixture 3
(38–94%, P = 0.0004), and Mixture 4 (69–100%, P < 0.005). The results of this study highlight
the potential limitations of dogs trained to detect a single odor to then recognize the odor
when mixed with other substances. The odor delivery device developed for this study rep-
resents an important and effective training option that may reduce the need for using a
final PC explosive mixture in canine training.
© 2013 Elsevier B.V. All rights reserved.

1. Introduction (Fischer-Tenhagen et al., 2011; Kiddy et al., 1978), and

It is well known that the domesticated dog possesses
highly developed olfactory abilities. Because of their keen
sense of smell, dogs are extensively used to detect a
broad variety of substances including narcotics (Adams
and Johnson, 1994; Dean, 1972), human remains (Lasseter
et al., 2003), cancers (Cornu et al., 2011; Pickel et al., 2004;
Walczak et al., 2012; Willis et al., 2004), cows in estrus
∗ Corresponding author at: North Carolina State University, College of
Veterinary Medicine, 1060 William Moore Drive, Raleigh, NC 27607, USA.
Tel.: +1 919 513 6058.
E-mail address: lucia lazarowski@ncsu.edu (L. Lazarowski).
bed bugs (Pfiester et al., 2008). Another well-established
role for dogs is scent detection of land mines, impro-
vised explosive devices (IEDs), undetonated munitions, and
other explosive materials that pose a risk to civilian and
military populations (Furton and Myers, 2001; Gazit and
Terkel, 2003; Harper et al., 2005; Jones, 2011). In addition to
effects on civilian populations, roadside bombs, suicide car
bombs, and other IEDs have caused the majority of Amer-
ican combat casualties in Iraq and Afghanistan (Wilson,
2007). Detection of these explosives by dogs and their ulti-
mate removal by trained personnel can reduce civilian and
military casualty rates and can reopen land for farming and
other purposes (Faust et al., 2011; Watts, 2009). Despite
the widespread use of dogs for scent detection, questions

0168-1591/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.applanim.2013.11.010
 L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93
remain about the underlying processes of explosive scent
detection in this and other species.
Training dogs to detect explosives presents several chal-
lenges. First, the types of explosives found in IEDs reflect
local availability and can vary widely from region to region.
Explosives commonly found in IEDs include organics
(e.g., 1,3,5-hexahydro-1,3,5-trinitrotriazine [RDX]; 2,4,6-
trinitrotoluene [TNT]), inorganic oxidizers (e.g., ammo-
nium nitrate [AN], potassium chlorate [PC]), or a com-
bination (e.g., Amatol – a RDX and AN mixture) (Kopp,
2008). Further, the use of homemade explosives (HMEs)
has recently become more common than commercial and
military explosives (Östmark et al., 2012). Consequently,
the absolute and relative amounts of explosive precur-
sors found in HMEs can vary widely. A second challenge
is presented when the base explosive is further modified
with additional gelling agents (e.g., wax or petroleum jelly),
fuels (e.g., diesel fuel or kerosene), or extraneous distract-
ing odors (Kopp, 2008). Therefore, most target (explosive)
odors encountered by dogs under field conditions are
comprised of a combination of many different substances
(Harper et al., 2005), which may differ from those used in
training. Learning to respond upon detection of a trained
odor, then, may not generalize to detection of novel odor
combinations. Thus, generalization from trained compo-
nents to novel configurations, such as mixtures composed
of trained and untrained odors, is an important feature
of canine scent detection. The concept of generalization,
in which an animal must efficiently and appropriately
respond to novel stimuli based on prior experience with
different stimuli, has been extensively studied (Ghirlanda
and Enquist, 2003; Rilling, 1977; Spence, 1937); how-
ever, olfactory generalization of compounds of military
interest in dogs remains largely unexplored (Johnston,
1999).
Generalization between a stimulus compound (e.g.,
a tone and a click presented together) and its ele-
ments generally occurs in one of two ways (Bouton
et al., 2012). Elemental, or analytical, processing occurs
when each element in the compound forms individ-
ual stimulus–response associations, which are retained
regardless of combining with other stimuli (Rescorla,
1972). In regards to olfactory processing, if combining two
or more odorants does not alter their individual proper-
ties such that the original odors remain identifiable in the
resulting mixture, conditioned responses to the individual
odors should generalize to the mixture (Linster and Smith,
1999). Alternatively, when conditioned stimuli are com-
bined the resulting compound may be perceived as a novel
configuration (Pearce, 1987), referred to as configural or
synthetic processing. Odorants may combine to create a
novel, blended mixture in which individual characteris-
tics of each odorant are altered, creating the perception
of a new and unique odor. In this case, conditioning to
an individual component of the mixture does not produce
generalization to a mixture that contains the component,
and vice versa (i.e., conditioning to a mixture does not
produce generalization to the individual components pre-
sented individually) (Derby et al., 1996; Laing and Francis,
1989; Staubli et al., 1987). The degree of generalization
from components to mixtures may vary depending on a
85
number of factors including the particular physicochemical
properties and identities of the odorants (Derby et al., 1996;
Kay et al., 2005; Laska and Hudson, 1993; Linster and Smith,
1999), the complexity of the mixture (Livermore and Laing,
1998a), the relative intensities of the odors (Livermore and
Laing, 1998b), and olfactory enrichment (Mandairon et al.,
2006).
Our study addressed several of these challenges. In our
experiment we explored whether dogs trained to detect
one component of an explosive will correctly detect a novel
mixture containing that component. Our second experi-
ment assessed the performance of dogs trained to reliably
detect the components of a mixture using a novel device
that maintained separation of the components in distinct
compartments while presumably producing a merged odor
presentation. The use of this device addresses the need to
develop training aids that can be handled safely and yet
provide the representative odor profile of the explosive of
interest.
2. Animals, materials and methods
2.1. Animals
Nine male and 11 female field-trial-bred Labrador
Retrievers ranging from 2 to 5 years in age were used.
All dogs were procured by a private military working dog
training firm (K2 Solutions, Southern Pines, NC, USA) from
field-trial-breeding kennels throughout the United States
(USA). Dogs were individually housed in 4.6 m × 1.5 m out-
door kennels at the K2 Solutions Canine Training Center
(K2) located in Southern Pines, NC, USA. This facility is
designed, equipped, and operated to comply with Title 9,
Code of Federal Regulations, parts 1–3 and with the USA
Department of Defense (DoD) Directive 3216.01 to guaran-
tee the humane, safe, and necessary use of dogs. Animals
were fed twice a day with commercial adult large breed dog
food (Purina ProPlan, St. Louis, MO, USA) and had unlimited
access to water. Experienced K2 trainers carried out the
training and handling of the dogs for the duration of the
study. All experiments were performed during a 5-week
period under ambient weather conditions. Maximum daily
temperatures ranged from 13 to 33 ◦ C (daily maximum
mean temperature = 27 ◦ C). All procedures were approved
by the North Carolina State University (NCSU) Institutional
Animal Care and Use Committee.
2.2. Odorants and containers
Odor delivery was accomplished using a custom-built
inverted “T”-shaped odor delivery device (Fig. 1). The odor
delivery device was constructed using commercially avail-
able 7.62 cm diameter polyvinyl chloride (PVC) pipe fittings
(JM Eagle, Los Angeles, CA, USA). The two terminal end-caps
of the “T” allowed for the placement of PC-based mate-
rials of interest (one or both arms were used depending
upon the experimental phase). Each terminal end-cap was
attached to a 45◦ elbow which was joined to a “tee” joint
fitting that terminated with a drain fitting and perforated
drain cover lid. Separate odor delivery devices were used
86 L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93
Fig. 1. Odor delivery device used in the experiments. Insert shows the
top view of the drain fitting as seen by the dogs. Odorants of interest were
placed in one or more end caps depending upon experimental phase. Air
containing the odorant(s) of interest passively diffused to the opening at
the top of the device (see Section 2.2 for more details).
for each odorant. Odorants used in this study included:
(a) chemically pure (98 to >99%) potassium chlorate (PC,
chemical formula KClO3 , Carolina Biologicals, Burlington,
NC, USA), (b) PC-based explosive mixtures that mimicked
materials used to produce HMEs in Afghanistan, (c) non-PC
sub-components (e.g., gelling and fueling agents) used to
create the PC-based explosive mixture, and (d) packaging
materials. The four PC-based explosive mixtures were syn-
thesized by chemists from the USA Naval Surface Warfare
Center by blending pure PC with one of the four non-
PC components. The identities of the non-PC components
used to produce the explosive mixtures were treated as
proprietary information. Various types of packaging con-
tainers were used throughout training and experimental
phases to contain the PC, components, and mixtures within
the odor delivery device including small plastic containers
(Tupperware Corporation, Orlando, FL, USA), glass con-
tainers with metal lids, plastic and cotton bags (similar
to materials used by Gazit and Terkel, 2003). The use of
several types of packaging materials is part of K2’s odor
training protocol intended to ensure that the dogs are
discriminating the target odors and not the odor of the
containers, and also mimics the use of various packaging
materials associated with IEDs. Odor delivery through the
device was static and relied upon diffusion of the odor-
ant through the device. Depending upon the experimental
phase, one or more odor delivery devices were used in the
test arena.
2.3. Experimental locations
All experimental testing was performed in a dedicated
outdoor test arena at the K2 facility. This purpose-built
floorless (earth and sand terrain) test arena had par-
tial wooden walls and a complete wooden ceiling (Home
Depot, Inc., Atlanta, GA, USA) and measured 7.3 m × 9.8 m
in size. The test arena’s 1.2 m tall wooden walls segregated
dog-trainer teams from the researchers and facilitated
observation of all events.
2.4. Experimental procedure
This study was performed in three phases: (a) Phase 1 –
initial PC training followed by a 10-trial assessment of the
dog’s ability to detect pure PC; (b) Phase 2 – tests evalu-
ating the ability of dogs to generalize from PC to PC-based
explosive mixtures; and (c) Phase 3 – evaluation of whether
training with the mixture components using a novel odor
delivery device improved the ability of dogs to detect PC-
based explosive mixtures. Phase 1 consisted of 10 trials
while Phases 2 and 3 were conducted in two separate 6-
trial sessions. All experimental phases were performed in
a sequence. Two trainers alternated handling the dogs on
each trial in order to facilitate testing. This, in combination
with randomizing the order of dogs, maximized dog inter-
actions with both handlers across all experimental phases.
Control trials in which odorant amounts were manipu-
lated were performed at the conclusion of the experimental
phases.
2.4.1. Phase 1 – initial PC training and assessment
Prior to beginning experimental test trials, all dogs
(n = 20) were trained by experienced K2 odor training staff
to detect PC while acclimating to the odor delivery devices.
Initial odor training with PC was performed at sites unasso-
ciated with the test arena used in subsequent experimental
trials. Dogs were trained to perform an off-leash search
and respond to approximately 450 g of PC placed inside
of the end cap of an odor delivery device by sniffing the
perforated tops and lying down in sternal recumbence in
proximity to the target odor delivery device. Upon demon-
stration of this response dogs were rewarded by trainers
through verbal reward and retrieval of a rubber Kong® toy
(KONG Company, Golden, CO, USA). The test arena held one
odor delivery device containing 450 g of PC inside one of the
end caps, as well as one or more (up to eight total, gradually
increasing across training sessions) visually identical con-
trol devices that contained either extraneous ‘distractor’
odors (e.g., soil, metal, tea bags, packaging materials) or
were left empty (‘blanks’). Blank trials were run occasion-
ally in which no PC was present in any of the odor delivery
devices. If a dog falsely responded to a device that did not
contain the target, the trainer ignored the dog and allowed
it to continue searching. Training sessions occurred
Monday–Friday for 18 days and varied day to day in the
number of search trials, pattern in which odor delivery
devices were presented (e.g., circular or line formations),
number of devices presented, and order of dogs run.
After initial PC training, a test was conducted in order
to validate that dogs could reliably detect and respond
to the trained amount of PC before proceeding with the
subsequent experimental phases. The training assessment
session consisted of a total of 10 trials: three test trials in
which the target odor (PC) was present, and seven trials
in which it was not. Odor delivery devices were placed
approximately 2 m apart in a circular formation inside the
test arena, and dogs were allowed to perform a search sim-
ilar to that in the previous training phase. Trials 1 and 7
were blank trials consisting of seven empty odor delivery
devices. On Trials 2, 4, and 6, a device that held approxi-
mately 450 g of PC was added, placed in a random position
 L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93
each time. Distractor odors were placed in the seven other
odor delivery devices. On Trials 3 and 5 only the seven
distractor odor delivery containers were present. Finally,
Trials 8–10 consisted of control trials to check for the possi-
bility of dogs responding based on the odor of the packaging
materials used to contain the target odor during training
instead of the PC odor. Three types of packaging materials
were tested, one trial at a time: a cotton bag, a plastic con-
tainer (Tupperware Corporation, Orlando, FL, USA), and a
glass container. On each of these trials, four odor delivery
devices were present, one of which contained the packag-
ing material and the other three were left empty.
All dogs were successively run through a trial before the
next trial type was set up. Prior to beginning the session,
a randomized list was generated to designate the order in
which the dogs were run. The order was re-randomized
after Trials 3 and 6. At the start of each trial, a dog was
walked to the testing arena on-leash by a K2 trainer and
positioned in the designated start area. The dog was then
removed from the leash and given a command to search.
During the search trainers remained at the arena entrance,
avoiding eye contact with the dog and maintaining neutral
body position in order to minimize unintentional cueing.
On target odor trials (2, 4, and 6), the dog was rewarded
with a Kong® toy upon displaying a correct response at the
target location. An additional trainer outside of the testing
arena and out of view of the dog tossed the toy into the
arena immediately after a correct response was exhibited.
The dog was then removed from the test arena while the
next dog was retrieved by a second trainer. Responses were
recorded as either positive (dog displayed response within
approximately 1 m of the device) or negative (dog searched
target device but failed to display correct response). Upon a
negative response, dogs were allowed to continue search-
ing the remainder of the devices in the arena and then
were removed with no reward. On non-target trials (1, 3, 5,
7–10), dogs were allowed to search the test arena and each
device sufficiently before being called back by the trainer.
Any responses to non-target devices were recorded as a
false response. False responses were ignored by trainers,
and the dog was allowed to continue searching the remain-
der of the test arena before being removed. In order to move
on to the next experimental phase, a criterion was required
of a positive response on all three of the target odor trials
and only dogs that met this criterion continued to the sub-
sequent experimental phase. False responses in this phase
were recorded and analyzed, but not used as exclusionary
criteria for moving to the next phase. In between subjects,
the odor delivery devices were wiped down with a cloth
and the test arena was lightly raked. After every fourth trial,
all odor delivery devices were removed from the area and
wiped down, and the entire test arena was raked. Three
data collectors observed from outside of the testing arena
and independently recorded trial information and results.
Additionally, all trials were videotaped.
2.4.2. Phase 2 – generalization tests
The goal of this phase was to evaluate whether dogs
that reliably detected pure PC in Phase 1 would respond
to a PC-based mixture (i.e., generalization of trained
responses to novel odor combinations). Each mixture was
87
produced on site by chemists associated with the USA
Naval Surface Warfare Center, Indian Head Division. Mix-
tures weighed approximately 200 g and were comprised
of 140–180 g of PC and 20–60 g of a non-PC component.
The amount of PC and non-PC component used in each
mixture varied depending upon the chemical composition
of the final explosive mixture tested. Because the iden-
tities of the non-PC components used to formulate the
final mixtures were to remain undisclosed, the PC-based
mixtures used in Phase 2 and beyond are designated as
Mixture 1, Mixture 2, Mixture 3, and Mixture 4. Dogs were
not previously exposed to the non-PC component at any
time.
Phase 2 was completed over 2 days, with one session
each day consisting of six trials each. Testing was per-
formed in sequence with all dogs. Eight odor delivery
devices were present in a circular formation on each trial.
On Trials 1 and 6, all of the odor delivery devices were
empty. On Trials 2 and 4, distractor odors were placed
inside each device. On Trials 3 and 5, one odor delivery
device, randomly chosen, was removed and replaced with
a device containing Mixture 1 (Trial 3) or Mixture 2 (Trial 5).
This sequence of trials was repeated on the second session
the next day except Trial 3 used Mixture 3, and Trial 5
used Mixture 4. In this phase, trials were conducted in
the same manner as the Phase 1 test trials (see Section
2.4.1), except the dogs were not rewarded for displaying
a positive response to the target container. Upon display-
ing a positive response the dogs were called out of the
test arena by the trainer who remained neutral. If a dog
did not indicate a response upon the first pass of the tar-
get device, the trial continued to allow the dog to search
the remainder of the devices before being called back by
the trainer. Responses were scored as positive (dog dis-
played immediate response within 1 m of target odor) or
negative (dog failed to respond to the target odor). False
responses (responding to a non-target device) were also
recorded. The testing order of dogs was randomized for
each mixture.
The first two trials of each session in this phase served as
exclusion criteria trials based on false responses. Dogs that
exhibited a false response on both Trials 1 and 2 of either of
the two sessions were removed from testing at this point.
This was implemented in order to eliminate dogs display-
ing high rates of false responding which could invalidate
their future test results.
2.4.3. Phase 3 – evaluation of novel training device
The purpose of this phase was to evaluate the effective-
ness of the odor delivery device as a training method for
odor presentation of binary mixture components (i.e., PC
and a second component). In this phase, PC was placed in
one end cap of the device, while the non-PC component
was placed in the remaining end cap. The design of the
device ensured that PC and the second component would
remain physically separated throughout the experiment;
however, the air from each is presumably combined by dif-
fusion and presented as a merged odor to the dog. Ratios
and amounts of the two components remained consistent
with the amounts used in the actual mixtures during the
test phase (see Section 2.4.2). In this phase, dogs received
88 L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93
training with the binary components that corresponded
to any mixtures in which they failed to respond to dur-
ing Phase 2. Dogs were trained to respond to the device
containing the separated mixture components using meth-
ods described earlier (see Phase 1, Section 2.3). Training
occurred Monday–Friday for 7 days. The criterion used to
confirm that odor training was successful was met when
the dog showed at least 10 consecutive unassisted positive
responses to the odor delivery device containing the target
odor.
Following training with the separated mixture com-
ponents, all dogs that advanced in Phase 2 (n = 16) were
re-tested on fresh preparations of the four PC-based mix-
tures. Testing occurred over two 6-trial sessions that were
conducted in the same manner as in Phase 2 (see Section
2.4.2). However in this phase, observers and trainers were
blind to the trial conditions (i.e., odor identity and loca-
tion).
2.4.4. Variation of odorant amount
Next, a session was conducted in which dogs were pre-
sented with the four mixtures containing novel amounts
of PC to determine the effect of varying the odorant quan-
tity on dogs’ performance. Dogs were tested with each
of the four PC-based mixtures containing increased and
decreased amounts of PC. Trials were run in the same
manner as the test sessions in Phases 2 and 3, except in
one session the amounts were increased to quantities ran-
ging from 300 to 700 g of PC, varying across mixtures, and
in the other session amounts were decreased to approxi-
mately 100–150 g of PC, varying across mixtures. Failure
to respond to mixtures containing novel amounts may
indicate that dogs learn to respond to the specific trained
amount of PC and may not recognize variations in odor
intensity.
Finally, additional trials were run in which each non-
PC component was presented in an odor delivery device
without the presence of any PC to determine if responding
was under control of the non-PC component alone, which
may have inadvertently developed during training. To test
this, one trial was performed for each non-PC component
in which a fresh sample of the non-PC component was
presented in a device along with five blank devices. Only
dogs that received the separated component odor merger
training in Phase 3 were tested.
2.5. Data analysis
Statistical analysis of categorical data (response vs. no
response to test odor or non-target devices) were ana-
lyzed using a nominal logistic model with a likelihood
ratio chi-square test (n = 30–32 [for 15 or 16 dogs], df = 1).
False response rates were calculated by dividing the sum
of false responses for all dogs combined by the total num-
ber of dogs tested for a given trial or phase. The impact
of handler effects on canine performance was assessed
using a Pearson Chi Square test. All statistical analy-
ses were performed using JMP Pro Version 9 (Cary, NC,
USA). The results were considered statistically significant
if P ≤ 0.05.
Fig. 2. Overall false response rates (%) for each type of trial (blanks, dis-
tractors, target odors) for Phases 1–3. False response rates were calculated
by dividing the sum of false responses for each trial type combined in a
phase by the total number of dogs tested.
3. Results
3.1. Initial PC training and assessment (Phase 1)
All dogs (n = 20) displayed a positive response on all
three test trials performed in Phase 1, satisfying the cri-
terion to advance to the next phase of the study (Table 1).
Twelve dogs displayed a false response on at least one of the
10 trials, but no more than three total false responses were
committed by one dog. The remaining eight dogs did not
display any false responses in the entire session. Two dogs
alerted to devices containing packaging material during
Trials 8–10 (note that these two dogs were later removed
from the study for excessive false responses during the
exclusion criterion trials of the subsequent phase). False
response rates were higher during distractor and blank tri-
als than target odor trials (Fig. 2), and highest on Trial 3
(Fig. 3).
A trial by trial analysis of dog performance during each
phase did not reveal any statistically significant differences
that could be attributed to an individual handler’s impact
on dog performance.
3.2. Generalization tests (Phase 2)
Four dogs were removed from the study based on the
false response exclusion criteria of Phase 2 (two consec-
utive false responses on Trials 1 and 2 of either session)
and the remaining dogs (n = 16) were tested on the four
PC-based mixtures. One dog did not participate in the first
session due to absence for medical reasons and therefore
was not tested on Mixtures 1 and 2.
Percentage of dogs that positively responded to each of
the four mixtures ranged from 27 to 69% across mixtures
(Table 2). Out of the 16 dogs tested, two showed a positive
response to all four mixtures (100% response rate); three
dogs showed a positive response to three of the four mix-
tures (75% response rate); four dogs had a 50% response
rate, and four dogs failed to show a positive response to
any of the PC-based mixtures responses (Table 1).
 L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93 89

Table 1
Individual percent correct for potassium chlorate (PC) detection assessment trials (Phase 1) and PC-based mixture generalization test trials (Phases 2 and
3). Individual total percentage of false responses for the entire session is also reported.

Dog name Phase 1: PC validation Phase 2: generalization test Phase 3: training evaluation

Percent correct Percent false Percent correct Percent false Percent correct Percent false

Annie 100 0 50 0 100 33

Audi 100 0 100 25 100 42
Brutus 100 10 0 17 100 0
Charlie 100 0 25 0 100 0
Cricket 100 0 0 0 100 0
Drake 100 20 0 17 100 8
Dye 100 0 50 8 75 0
Frieda 100 10 nda nda nda nda
Green 100 10 nda nda nda nda
Harley 100 10 100b 17 100 8
Heidi 100 30 nda nda nda nda
Ike 100 0 25 0 100 0
Indy 100 10 50 0 100 8
Keppie 100 30 nda nda nda nda
Kody 100 30 100 17 100 0
Moss 100 0 75 0 100 0
Pirate 100 0 50 17 50 8
Sadie 100 10 75 17 100 0
Sandy 100 20 0 8 100 33
Twiggy 100 10 75 0 75 0
a
No data (nd) indicates dogs eliminated due to false responses.
b
Score is out of two trials rather than four due to this dog’s absence during one session for medical reasons.

Individual false response rates over the 12 trials ranged
from 0 to 25%, with seven of the 16 dogs displaying no false
responses in the entire session (Table 1). False response
rates were highest during distractor trials (Fig. 2) and on
Trial 2 (Fig. 3). Table 2 shows individual dogs’ results for
each of the PC-based mixture test trials for the Phase 2
generalization test (‘pre’).
Fig. 3. Overall false response rates (%) as a function of trial number for
Phases 1–3. False response rates were calculated by dividing the sum of
false responses for each trial in a phase by the total number of dogs tested
in the phase. Phase 1 assessed detection of pure potassium chlorate (PC)
and was completed over 10 consecutive trials, with PC tested on trials
2, 4, and 6. Phases 2 and 3 assessed detection of PC-based mixtures and
were performed in two separate 6-trial sessions, with mixtures tested on
trials 3 and 5. Trial data for the two sessions in each of Phases 2 and 3 are
combined for that phase.
3.3. Evaluation of novel training device (Phase 3)
Next, dogs (n = 16) received additional training corre-
sponding to individual performance during each of the
Phase 2 generalization tests for 7 days until a criterion of
10 consecutive positive responses per odor was achieved.
The two dogs (‘Audi’ and ‘Kody’) with 100% response accu-
racy in Phase 2 were included in the remaining phases as
controls. All dogs (n = 16) were then re-tested on the four
PC-based mixtures. Total percentage of dogs correctly res-
ponding to each mixture ranged from 81 to 100% across the
four mixtures (Table 2). Positive responses to the PC-based
mixtures were significantly higher than responses to the
PC-based mixtures in Phase 2 (Fig. 4); significant improve-
ments were seen with Mixture 1 (2 = 22.03, P < 0.0001,
n = 30, df = 1), Mixture 2 (2 = 5.18, P = 0.0229, n = 30, df = 1),
Mixture 3 (2 = 12.53, P = 0.0004, n = 32, df = 1), and Mix-
ture 4 (2 = 7.86, P < 0.005, n = 32, df = 1). Thirteen of the 16
dogs showed a positive response to all four mixtures (100%
response rate); two dogs showed a positive response to
three of the four mixtures (75% response rate), and one dog
performed with a 50% response rate (Table 1). Individual
dogs’ responses to each of the four mixtures before (‘pre’)
and after (‘post’) Phase 3 odor merger device training are
reported in Table 2.
In Phase 3, individual false response rates by dogs
ranged from 0 to 42% over the 12 trials, with nine of the
16 dogs showing no false responses in the entire session
(Table 2). False response rates were highest during dis-
tractor trials (Fig. 2) and on Trial 2 (Fig. 3).
3.4. Variation of odorant quantities
All dogs tested demonstrated a positive response on
each of the eight trials (100% response rate) in which
90 L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93

Individual responses to each potassium chlorate (PC)-based mixture after initial odor training with chemically pure PC (‘pre’), after training with an odor delivery device that contained the physically segregated
PC and non-PC components (‘post’), to the four non-PC mixture components presented without PC (‘0-PC’), and to mixtures containing increased (‘+PC’) and decreased (‘−PC’) amounts of PC. Individual responses

−PC

100
nda

nda
nda
nda
nda

nda
nda
nda

nda
1
1
1
1
1

100
+PC
nda

nda
nda
nda
nda

nda
nda
nda

nda
1
1
1
1
1

1
0-PC
nda

nda
nda
nda
nda

nda
nda
nda

nda
14
1
0
0
0
0
0

0
Post
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
100
Mixture 4

Pre
1
1
0
0
0
0
1
1
1
1
1
1
1
1
0
1
69
−PC

100
nda
nda

nda

nda
1
1
1
1
1
1

1
1

1
1
1
100 Fig. 4. Percentage of dogs demonstrating a positive response to the four
+PC

nda
nda

nda

nda

potassium chlorate (PC)-based explosive mixtures. Data were collected

1
1
1
1
1
1

1
1

1
1
1

following training with either PC alone in Phase 2 (denoted here as Test 1)

or following training with the separated mixture components contained
0-PC

in the odor delivery device in Phase 3 (Test 2).

nda
nda

nda

nda

33
1

1
1
0
0
0
0
0

varying odorant amounts were used (Table 2). Seven of the

Post

16 dogs showed a positive response to at least one of the

1
1
1
1
1
1
1
1
1
1
1
1
0
1
1
1
94
Mixture 3

four non-PC mixture components when presented alone

(Table 2). Total percentage of positive responses across the
Pre
0
1
0
0
0
0
1
1
0
1
1
0
1
0
0
0
38

four non-PC components ranged from 0 to 36%.

−PC

100
nda

nda

nda

4. Discussion
1
1
1
1
1
1
1
1
1
1

1
1

The first main objective of our study was to determine

100
+PC
nda

nda

nda

whether dogs trained to detect the scent of chemically pure

1
1
1
1
1
1
1
1
1
1

1
1

PC could then generalize this behavior when presented

with different PC-based explosive mixtures. We found that
0-PC
nda

nda

nda

nda

most dogs trained with pure PC did not correctly signal

0
0

0
0
0
0
0
0
0

0
0
0

the presence of one or more PC-based explosive mixtures,

despite having demonstrated the ability to reliably detect
Post
1
1
1
1
1
1
0
1
1
1
1
1
0
1
1
0
81

a trained amount of pure PC. Our research group has seen a

Mixture 2

No data (nd) due to exemption based on performance in earlier phases.

similar inability by dogs to generalize from pure ammo-

nium nitrate to ammonium nitrate-based explosives in
ndb
Pre

40
1
1

1
1

1
0

0
0
0

0
0

0
0

controlled laboratory-based olfactory discrimination stud-

ies (Dorman unpublished observations).
−PC

Several possibilities may explain the failure to detect a

100
nda

nda

No data (nd) due to dog’s absence for medical reasons.

1
1
1
1
1
1
1
1
1
1
1

1
1

previously trained substance when presented as part of a

novel mixture of odors. Detection of individual components
100
+PC

nda

nda

within an olfactory mixture may depend on the manner in

1
1
1
1
1
1
1
1
1
1
1

1
1

which they interact with each other, which largely deter-

mines whether the physical properties are maintained and
0-PC

nda

nda

recognizable in the resulting mixture (Derby et al., 1996).

36
scored as positive (‘1 ) or negative (‘0 ).

1
1

1
1

1
0
0
0
0
0
0
0

Mixture suppression is a common interaction that occurs

when the response to a blend of odorants is less than would
Post
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
100

be predicted by the cumulative addition of responses to the

Mixture 1

individual odorants that comprise the final mixture. Inter-

actions may depend on the specific type of odorants in the
ndb
Pre

27

mixture, and may result from components suppressing or

1

1
1

1
0

0
0
0
0
0

0
0

0
0

altering each other (Derby et al., 1985; Linster and Smith,

1999). In the present study, combining the precursor com-
Dog name

Total (%)

ponents to form the explosive material may have caused

Twiggy
Charlie
Cricket

Harley
Brutus

Sandy
Drake
Table 2

Annie

Pirate
Sadie
Moss
Kody
Audi

Indy
Dye

chemical or physical interactions in a way that changed the

Ike

sensory perception of the individual components. Thus, it

 L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93
is possible that the lack of responses to PC-based mixtures
by dogs that were proficient in detecting pure PC was due
to changes in the perception of PC caused by its interac-
tion with the non-PC component. This may also explain
the variation seen between responsiveness to each of the
four mixtures, in that the specific components in each mix-
ture may have interacted with PC in different ways and
to differing degrees. For example, we found that Mixture
1 was associated with the lowest percentage of positive
responses (25%), indicating that the non-PC component
used in this mixture may be more reactive with PC than
other non-PC components tested. In contrast, Mixture 4
was associated with the highest positive response rate
(71%) suggesting that the non-PC component in this mix-
ture was either more inert or produced minimal changes
to the odor properties of pure PC. However, given that the
first mixture tested received the lowest response rate and
the last mixture tested received the highest, it is possible
that order effects were also a contributing factor.
Our findings are somewhat surprising since previous
studies have shown that dogs are capable of discriminat-
ing a target odorant when mixed with various amounts of
another odorant (Johnston, 1999). However, previous stud-
ies held amounts of the trained target substance constant
throughout tests in which it was mixed with other odors.
In the current study, dogs were initially trained with 450 g
of PC but test mixtures contained reduced amounts of PC
(140–180 g) due to the explosive properties of the mix-
ture. It is possible then that the lack of responses to the
PC-based mixtures in Phase 2 was due to the variation in
the amount of PC present in the mixture compared to the
amount used in training. One possible effect of reducing the
overall amount of PC is that the intensity of the odor was
also reduced. Changes in stimulus intensity are known to
produce changes in responding, sometimes referred to as
stimulus intensity dynamism (Hull, 1949), an effect that is
unique to stimulus intensity and differs from other stim-
ulus dimensions such as wavelength, size, and frequency
(Mackintosh, 1974). In classic studies of Pavlovian general-
ization as well as instrumental generalization, increasing
the intensity of the conditioned stimulus produced an
increase in responding; similarly, a reduction in respon-
ding was seen to stimuli less intense than the conditioned
stimulus (see Mackintosh, 1974 for review).
In order to evaluate the possibility that varying the
amount of odorant used could cause the dogs to fail to
generalize from standard training amounts to novel quan-
tities, additional trials were performed with increased and
decreased amounts of PC. We found that neither increasing
nor decreasing the amount of PC in the mixtures affected
the dogs’ ability to recognize the target odor and all dogs
responded to each of the eight novel variations presented. It
is possible, however, that these variations were not robust
enough to cause a disruption in detection. It is also pos-
sible that further training and experience with the odors
during Phase 3 facilitated the dogs’ ability to generalize to
the variations. However, a fixed amount of PC was used
during training (corresponding to the ratios used in mix-
tures during test trials) and the positive responses to the
novel variations suggests that the failure to generalize to
the PC-based mixtures was not likely due to the change in
91
amount of PC, although this cannot be entirely ruled out.
Nevertheless, our finding that dogs demonstrating profi-
ciency in detecting a trained amount of pure PC did not
reliably generalize to the PC-based mixtures, whether due
to changes in amount of PC or interaction with fueling
agents, is important and a further understanding of what
factors may facilitate generalization is needed.
It is also important to recognize that differences in
individual animal responses exist despite training. For
example, we found that less than 50% of the dogs trained
to pure PC subsequently responded to either Mixture 2
or 3. This finding suggests that perception of these mix-
tures varied greatly between dogs. Moreover, some dogs
demonstrated a 100% response rate to the four mixtures,
while some dogs failed to respond to all or most of the mix-
tures. Differences in individual responsiveness, sensitivity,
motivation, and experience may contribute to individ-
ual differences in odor mixture discrimination (Laska and
Hudson, 1993; Rabin and Cain, 1990). While training with
PC alone appeared to be sufficient to produce generaliza-
tion to PC-based mixtures for some dogs, this method did
not produce reliable detection in all dogs that otherwise
consistently detected trained amounts of pure PC.
Our second main objective was to determine whether
training the dogs with a blended odor that was developed
from the separated components could enhance detection
of the physically mixed PC and non-PC components. To
this end, we developed an odor delivery device that safely
contained the PC and non-PC components but kept the
materials physically segregated while presenting a combi-
nation of the two odorant air streams to the dog. The odor
delivery device was composed of commercially available
PVC pipe fittings and could be easily constructed and used.
We found that training with the separated components
contained within the device significantly improved the
positive detection rate for PC-based explosives. After train-
ing with the separated components in the device, overall
positive response rates to the four PC-based explosive mix-
tures increased from 27–69% to 81–100%. Individual dog
performance also became more uniform. During the first
generalization test (Phase 2), only two of the 16 dogs
responded with 100% accuracy. Of the remaining dogs
that scored below 100% on the first test and proceeded to
receive further training using the odor delivery device dur-
ing Phase 3, accuracy increased upon re-test of the mixtures
for 12 of the 14 dogs.
Overall, these results suggest that in this case, exclu-
sively training dogs with a standard amount of a single
target substance was not sufficient to produce reliable gen-
eralization to mixtures containing that substance, and that
in order to obtain high levels of positive detection of the
target binary mixtures, dogs needed to be trained with
both of the components. However, results from the Phase
3 trials in which dogs were trained to the segregated com-
ponents and subsequently tested on the explosive mixture
indicated that the two odorants did not need to be physi-
cally mixed during training in order to produce subsequent
generalization to the mixture.
Training and testing of the dogs was constrained by
certain standard operating procedures used by the K2 facil-
ity that were beyond the investigators’ control. First, for
92 L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93
safety reasons, trainers were not blinded to the location
of the target device, presenting the possibility of uninten-
tional cueing by gaze, posture, and proximity to the dog or
odor (Lit et al., 2011). However, dogs in the current study
were always worked off-leash and handlers remained at a
distance while avoiding eye contact and maintaining neu-
tral posture. The observers and data collectors remained
outside of the test arena, separated by wooden walls that
obstructed the dog’s view of the people from the chest
down. Throughout the study, dogs did not appear to attend
to any of the people outside of the testing area as they were
largely out of view. This is likely due to training practices
at K2 in which dogs are trained to attend exclusively to the
target objects and odors while ignoring extraneous sources
of distraction including other people, animals, and sounds.
Further, the low response rates seen in Phase 2 generaliza-
tion test trials that occurred despite observers’ and trainers’
awareness of the target location suggests that the dogs did
not use human directed cues to locate the target. Addi-
tionally, the majority of false responses occurred on trials
in which observers knew that no target odor was present
(Fig. 2). Although it cannot be ruled out that some dogs used
unintentional trainer or observer cueing, we feel that it did
not play a significant role in the dogs’ performance. How-
ever, efforts should be made in future studies to minimize
such a possibility.
It may be argued that false responses seen in several of
the dogs throughout testing may weaken the interpreta-
tion of the positive responses seen. However, overall false
response rates throughout the study were relatively low. In
Phase 1, false responses were recorded but were not used
as exclusionary criteria. In this phase, dogs simply needed
to demonstrate proficiency in reliably detecting PC. False
responses at this stage were likely due to the novelty of the
training paradigm. Dogs were still early in their training
and habituating to searching the devices in the test arena.
At the start of Phase 2, four dogs were excluded based on
displaying an excessive number of false responses (defined
as two or more in Trials 1 and 2 of either session) in order to
minimize potential confounding of test data. Of the 16 dogs
that advanced, five dogs never exhibited a false response in
subsequent phases of the study. The remainder of the dogs
made an average of 18% false responses in Phases 2 and 3
combined.
Few studies have systematically investigated causes
for false responses in detection dogs other than those
induced by handler cueing (Lit et al., 2011). However, false
responses may occur for a variety of other reasons. One
suggested cause of false responding is that dogs may learn
to respond based on learned temporal patterns of rein-
forcement. Although blank searches were programmed
into early training in Phase 1, dogs encountered the tar-
get odor, and thus reinforcement, more frequently. Typical
early training protocols involve numerous training prob-
lems in which dogs will encounter either continuous or
frequent intermittent reinforcement in a short period of
time in order to maintain attention and responding. As
a result, dogs may become habituated to the pattern of
repetitive reinforcement and may learn to anticipate rein-
forcement during a given timeframe (Herstik, 2010). This
hypothesis may explain the rise in false responses seen
from Trial 1 (blank trial) to Trial 2 (distractor trial) in Phases
2 and 3 (Fig. 3). Phase 1 consisted of a pattern of alternat-
ing non-target and target trials, which the dogs may have
learned. In the following phases, the first target odor trial
did not appear until the third trial, and thus dogs may have
anticipated an odor trial on Trial 2. However, given this type
of temporal based responding, false responses would also
be expected to occur during the final four non-rewarded
trials of Phase 1, in which no false response were recorded.
The majority of false responses occurred during dis-
tractor trials, and rarely occurred during target odor or
blank trials (Fig. 2). This suggests that another possible
cause for false responses in this study was that the dis-
tractor odors elicited responses. Items used as distractors
may have been composed of materials similar to the target
odor or in previous training the dogs received prior to this
study. Finally, although care was taken to clean odor con-
tainers and containers were never used for more than one
odorant, the possibility still exists that some cross contam-
ination may have occurred during handling of containers.
False responding may also simply be an artifact of train-
ing. During initial PC training (Phase 1), dogs were not
rewarded on blank trials for correctly rejecting (not respon-
ding to) the non-target containers, nor was a punishment
procedure ever used to correct a false response. Thus, the
relative response cost of responding (lying down) at a non-
target container was very low, and some dogs may have
adopted this behavioral strategy. Further, dogs were never
rewarded during Phase 2 or test sessions in Phase 3 in order
to prevent within-session learning that could affect inter-
pretation of any generalization seen. Thus, false responses
could be interpreted as extinction effects in which the
trained behavior initially increases when reinforcement for
a previously rewarded response is terminated (Lerman and
Iwata, 1995). Indeed, false responses were higher in Phases
2 and 3 than in Phase 1 (Fig. 2).
Other trials demonstrated that approximately 50% of the
tested dogs showed a positive response to at least one of
the non-PC components when presented alone. Positive
responses across individual non-PC components ranged
from 14 to 33%. These results indicate that for some dogs,
responding may have been controlled by the non-PC com-
ponent alone and was independent of the presence of PC,
a possible artifact of the previous training phase in which
the non-PC component was always part of the odor dis-
crimination. However, this is not conclusive and responses
during these trials may be due to reasons previously men-
tioned. Although not implemented during this study, the
use of the non-PC components as distractors during train-
ing and a differential reinforcement procedure may have
minimized this effect.
5. Conclusion
The results from the present study demonstrate that
training dogs to detect PC in its pure form may not be suf-
ficient to produce reliable, operational levels of detection
of PC-based explosive mixtures, although some individ-
ual exceptions were seen. This is an important finding
with implications to canine scent detection training in that
single-odor training, which is commonly used, may not
 L. Lazarowski, D.C. Dorman / Applied Animal Behaviour Science 151 (2014) 84–93
produce adequately trained dogs able to detect complex
explosive mixtures. Future studies investigating factors
that facilitate olfactory generalization would have impor-
tant implications in the training of scent detection dogs.
We also found that training dogs with a novel odor delivery
device, which produced an odor mixture from segregated
components, improved generalization to odor mixtures.
Despite certain limitations, this device represents a safe
and effective training option that reduces the need for using
a final PC-based explosive mixture in canine training.
Acknowledgements
This work was funded through a contract to K2 Solu-
tions, Inc. from the United States Office of Naval Research
(ONR). Lisa Albuquerque, Michael Hoglund and other mem-
bers of the ONR provided logistical and technical support.
Lisa Albuquerque also designed and constructed the odor
delivery device and holds a provisional patent (61/814,385)
for the device. We thank the USA Naval Surface Warfare
Center, Indian Head Division and K2 Solutions personnel
for facilitating this study as well as Joseph Albuquerque for
his input on technical matters.
References
Adams, G.J., Johnson, K.G., 1994. Sleep, work, and the effects of shift work
in drug detector dogs Canis familiaris. Appl. Anim. Behav. Sci. 41,
115–126.
Bouton, M.E., Doyle-Burr, C., Vurbic, D., 2012. Asymmetrical generaliza-
tion of conditioning and extinction from compound to element and
element to compound. J. Exp. Psychol. Anim. Behav. Process. 38 (4),
381–393.
Cornu, J.N., Cancel-Tassin, G., Ondet, V., Girardet, C., Cussenot, O., 2011.
Olfactory detection of prostate cancer by dogs sniffing urine: a step
forward in early diagnosis. Eur. Urol. 59 (2), 197–201.
Dean, E.E., 1972. Training dogs for narcotic detection. Technical report No.
LWL-CR-6ODJ71. http://www.dtic.mil/cgi-bin/GetTRDoc?Location=
U2&doc=GetTRDoc.pdf&AD=AD0749302 (5.02.13).
Derby, C.D., Ache, B.W., Kennel, E.W., 1985. Mixture suppression in olfac-
tion: electrophysiological evaluation of the contribution of peripheral
and central neural components. Chem. Senses 10, 301–316.
Derby, C.D., Hutson, M., Livermore, B.A., Lynn, W.H., 1996. Generaliza-
tion among related complex odorant mixtures and their components:
analysis of olfactory perception in the spiny lobster. Physiol. Behav.
60 (1), 87–95.
Faust, A.A., de Ruiter, C.J., Ehlerding, A., McFee, J.E., Svinsas, E., van Rhee-
nen, A.D., 2011. Observations on military exploitation of explosives
detection technologies. In: Proc. SPIE 8017, Detection and Sensing of
Mines, Explosive Objects, and Obscured Targets XVI, 801716, May 23,
2011, http://dx.doi.org/10.1117/12.886391.
Fischer-Tenhagen, C., Wetterholm, L., Tenhagen, B.A., Heuwieser, W.,
2011. Training dogs on a scent platform for oestrus detection in cows.
Appl. Anim. Behav. Sci. 131, 63–70.
Furton, K.G., Myers, L.J., 2001. The scientific foundation and efficacy of
the use of canines as chemical detectors for explosives. Talanta 54,
487–500.
Gazit, I., Terkel, J., 2003. Explosives detection by sniffer dogs following
strenuous physical activity. Appl. Anim. Behav. Sci. 81, 149–161.
Ghirlanda, S., Enquist, M., 2003. A century of generalization. Anim. Behav.
66, 15–36.
Harper, R.J., Almirall, J.R., Furton, K.G., 2005. Identification of dominant
odor chemicals emanating from explosives for use in developing
optimal training aid combinations and mimics for canine detection.
Talanta 67, 313–327.
Herstik, M., Herstik, M., 2010. The false alert: reasons and remedies. Det-
onator 37, 43–45.
93
Hull, C.L., 1949. Stimulus intensity dynamism (V) and stimulus general-
ization. Psychol. Rev. 56, 67–76.
Johnston, J.M., 1999. Canine Detection Capabilities: Operational Impli-
cations of Recent R&D Findings. Institute for Biological Detection
Systems, Auburn University, Auburn, AL.
Jones, B.M., 2011. Applied behavior analysis is ideal for the development
of a land mine detection technology using animals. Behav. Anal. 34,
55–73.
Kay, L., Crk, T., Thorngate, J., 2005. A redefinition of odor mixture quality.
Behav. Neurosci. 119, 726–733.
Kiddy, C.A., Mitchell, D.S., Bolt, D.J., Hawk, H.W., 1978. Detection of estrus-
related odors in cows by trained dogs. Biol. Reprod. 19, 389–395.
Kopp, C., 2008. Technology of improvised explosive devices. Defense
Today, 46–49.
Laing, D.G., Francis, G.W., 1989. The capacity of humans to identify odors
in mixtures. Physiol. Behav. 46, 809–814.
Laska, M., Hudson, R., 1993. Discriminating parts from the whole: determi-
nants of odor mixture perception in squirrel monkeys, Saimiri sciureus.
J. Comp. Physiol. A 173 (2), 249–256.
Lasseter, A.E., Jacobi, K.P., Farley, R., Hensel, L., 2003. Cadaver dog and
handler team capabilities in the recovery of buried human remains in
the southeastern United States. J. Forensic Sci. 48 (3), 617–621.
Lerman, D.C., Iwata, B.A., 1995. Prevalence of the extinction burst and its
attenuation during treatment. J. Appl. Behav. Anal. 28 (1), 93–94.
Linster, C., Smith, B.H., 1999. Generalization between binary odor mixtures
and their components in the rat. Physiol. Behav. 66 (4), 701–707.
Lit, L., Schweitzer, J.B., Oberbauer, A.M., 2011. Handler beliefs affect scent
detection dog outcomes. Anim. Cogn. 14 (3), 387–394.
Livermore, A., Laing, D.G., 1998a. The influence of chemical complexity
on the perception of multicomponent odor mixtures. Percept. Psy-
chophys. 60 (4), 650–661.
Livermore, A., Laing, D.G., 1998b. The influence of odor type on the dis-
crimination and identification of odorants in multicomponent odor
mixtures. Physiol. Behav. 65 (2), 311–320.
Mackintosh, N.J., 1974. The Psychology of Animal Learning. Academic
Press, London/New York.
Mandairon, N., Stack, C., Linster, C., 2006. Olfactory enrichment improves
the recognition of individual components in mixtures. Physiol. Behav.
89, 379–384.
Östmark, H., Wallin, S., Ang, H.G., 2012. Vapor pressure of explosives: a
critical review. Propell. Explos. Pyrot. 37, 12–23.
Pearce, J.M., 1987. A model for stimulus generalization in Pavlovian con-
ditioning. Psychol. Rev. 94, 61–73.
Pfiester, M., Koehler, P.G., Pereira, R.M., 2008. Ability of bed bug-detecting
canines to locate live bed bugs and viable bed bug eggs. J. Econ. Ento-
mol. 101 (4), 1389–1396.
Pickel, D., Manucy, G.P., Walker, D.B., Hall, S.B., Walker, J.C., 2004. Evidence
for canine olfactory detection of melanoma. Appl. Anim. Behav. Sci. 89,
107–116.
Rabin, M.D., Cain, W.S., 1990. Attention and learning in the perception of
odor mixtures. In: Laing, D.G., Cain, W.S. (Eds.), Perception of Complex
Smells and Tastes. Academic Press, Sydney, pp. 173–188.
Rescorla, R.A., 1972. “Configural” conditioning in discrete-trial bar press-
ing. J. Comp. Physiol. Psychol. 79, 307–317.
Rilling, M., 1977. Stimulus control and inhibitory processes. In: Honig, L.,
Staddon, W.K.J.E.R. (Eds.), Handbook of Operant Behavior. Prentice-
Hall, Englewood Cliffs, NJ, pp. 432–480.
Spence, K., 1937. The differential response in animals to stimuli varying
within a single dimension. Psychol. Rev. 44, 430–444.
Staubli, U., Fraser, D., Faraday, R., Lynch, G., 1987. Olfaction and the “data”
memory system in rats. Behav. Neurosci. 101, 757–765.
Walczak, M., Jezierski, T., Gorecka-Bruzda, A., Sobczynska, M., Ensminger,
J., 2012. Impact of individual training parameters and manner of taking
breath odor samples on the reliability of canines as cancer screeners.
J. Vet. Behav. 7, 283–294.
Watts, H.G., 2009. The consequences for children of explosive rem-
nants of war: Land mines, unexploded ordnance, improvised
explosive devices, and cluster bombs. J. Pediatr. Rehabil. Med. 2,
217–227.
Willis, C.M., Church, S.M., Guest, C.M., Cook, W.A., McCarthy, N., Bransbury,
A.J., Church, M.R.T., Church, J.C.T., 2004. Olfactory detection of human
bladder cancer by dogs: proof of principle study. Br. Med. J. 329 (7468),
7.
Wilson, C., 2007. Improvised explosive devices (IEDs) in Iraq and
Afghanistan: Effects and countermeasures. CRS Report for Congress.
http://www.fas.org/sgp/crs/weapons/RS22330.pdf (accessed
5.02.13).