replacing it by its opposite, the belief is not justified. If the set is more coherentwith this belief as a member rather than with any alternative, the belief is justified. The link between justification and truth here is that the increasing coherence of a belief-set is, by definition, an increase of the justification of its members, and bydefinition the unique coherent set of propositions which account for the data of experience is the sum of all truth. As one's belief set increases in coherence, one has better and better reason to think one's beliefs true. Indeed, as Dancy points out, it isdifficult to distinguish between the coherence criterion of truth and a coherence definitionof justification. The reason this is so appealing is that without the coherence account of truth thecoherence theory of justification is not obviously truth-directed. Why should an increasein the mutual explanatory relations of my beliefs be taken as a sign, even a probable sign,that my beliefs are true? On the minimal correspondence theory of truth sketched earlier,there is no obvious answer to this question. But Dancy's marriage of coherence theories of truth and justification is not without problems. In the first place, the introduction of observational control on the relevantcoherent set of beliefs seems to turn this into a version of moderate foundationalism. Theaim of thought and inquiry on this view is to start from the data of experience and to construct a set of beliefs around thosedata which will order the data in the most systematic (coherent) way. To do thiswe may need to reject some of the data, but we cannot reject them all because our very aim is to make sense of what we have as data. So the set of beliefs which wedo construct must be empirically grounded, and this grounding in the data of experience guarantees that there will be only one set which constitutes 'the mostsystematic ordering.' Again, waiving questions as to whether this really does define ‘truth' in a way that isunique, and waiving as well worries as to whether any definition which links truth sotightly to subjective perceptual data can retain the notion of truth as non-epistemic, thedescription here of the human cognitive enterprise should warm the heart of anymoderate foundationalist. The data of experience form a privileged set, and we cannotreject them all, or apparently even a very high proportion of them, since it is our goal togive an account of them. Other beliefs which help us to give a better explanatory accountof them (
) are justified in virtue of this fact. But in this case, the privilegedstatus of the observational beliefs requires some account independent of their explanatory powers, an account of their favorable epistemic status which is separate from the notionof justification Dancy is wielding. The definition of justification as explanatorycoherence has apparently led him to a foundationalist theory of epistemic justification,and that in turn requires a notion of justification that goes beyond the bounds of thecoherence concept.