This paper offers a DRT-based analysis of epistemic specificity. Following Farkas (1996), we distinguish between scopal, partitive and epistemic specificity. After arguing in Section 1 that the three main variants of specificity are irreducible to each other, the paper then focuses on epistemic specificity. In the analysis of epistemically specific indefinites we distinguish between specific use and specific interpretation. Specific use is defined as a relation between (the semantic representation of) a linguistic form and (the representation of) the speaker's mental state: In the speaker's state the sentence containing the relevant indefinite corresponds to a singular proposition. Specific interpretation is in a sense a derivative notion: It characterises the representation constructed by the hearer just in case he construes an indefinite as having been used specifically by the speaker, and builds his own representation accordingly. The representation language we employ is a descendant of the original DRT-language presented in Kamp and Reyle (1993). This framework is tailor-made for the representations of attitudes of cognitive agents (for a recent discussion see Kamp (2013)); in the analysis reported in the present paper it enables us to distinguish between (i) the representation of an utterance that is derived via (standard) linguistic analysis, and (ii)-(iii) the representations that the individual discourse participants have or construct for this utterance. The key concept of the analysis is the notion of an anchored entity representation: Anchored entity representations are constituents of mental states that are causally linked, via their anchors, to the entities that they represent. In general, when a speaker uses a noun phrase to refer to an entity represented by one of her entity representations and thereby activates an entity representation in the mind of the hearer, the anchor of the hearer's representation, ERH , will often be structurally different from that of the speaker's own entity representation ERS, although normally the two representations will be coreferential. There will be a structural difference in particular when the speaker refers to the entity represented by ERS through making a specific use of an indefinite noun phrase. If the hearer takes her to have used the indefinite specifically, he will construct an entity representation ERH whose anchor links it to its referent as the entity represented by the speaker's representation ERS. Anchors of this type are called 'vicarious anchors'. The paper concludes with a brief discussion of the linguistic status of epistemic specificity. Data from English have been taken to suggest that specificity is an epiphenomenon, viz. that it need not be captured at the level of grammatical representation. But data from Romanian appear to suggest otherwise: The behaviour of Romanian indefinites marked with the Accusative preposition pe suggests that the specificity properties of these indefinites need to be marked at the level of (compositional) semantics, vi...