What happens if one treats social identity as a flexible resource in conversational interaction? Close attention to the sequencing of talk suggests that speakers' identities are much more subtle than simple pre‐given category labels suggest, and that they change rapidly as a function of the ephemeral (but socially consequential) demands of the situation. Were a psychologist to have sampled the interaction only at one given point, they would have seen a participant using, or being attributed with, only one identity; but we show that speakers use, and attribute each other with, a variety of different identities as their business progresses. In so doing, the speakers can be seen not only to avow contradictory identities but also to invoke both group distinctiveness and similarity—and neither of these strategies are easy to square with social psychological theories of identity. We put what we find in this particular case study into the debate between, on the one hand, ethnomethodological preference for working from participants' own orientations to identity and, on the other hand, social psychological research practices which tend to privilege analytically given social categories. At the very least, we argue, the social psychological approach can be enriched by attending more to identity as a matter of situated description and less as a matter of perceptuo‐cognitive processing.