This paper explores the notion of truthfulness in research on conversational remembering. It argues that people's accounts of past events, before they can be taken as data on the cognitive workings of memory, need to be examined as contextualized and variable productions that perform pragmatic and rhetorical work; no one version can be taken as a person's real memory. The consequences of this discourse-analytical perspective are examined first through a critical discussion of Ulric Neisser's study of John Dean's testimony to the senate ‘Watergate’ committee. The issues are then explored more deeply in an analysis of reportings of a different event, in which similar (Watergate-like) issues of memory, truth and accuracy are also at issue. It is argued that: (a) all of Neisser's three kinds of memory–verbatim, gist and ‘repisodic’–involve problematical assumptions concerning their relation to some true, original event; and (b) that Dean's accounts of his memory and his displays of memory should be approached as occasioned productions oriented pragmatically to the assignment of guilt and avoidance of scapegoating. Through an analysis of newspaper reports (based on memory) of a controversial briefing given by the then British Chancellor of the Exchequer, Nigel Lawson, we then examine: (a) how discourse about what could be used as an arbiter of truth was rhetorically organized; (b) how participants' versions of events were constructed rhetorically, as parts of arguments; (c) that both sides in the dispute maintained the coherence of their positions by a form of error accounting similar to that used by Neisser with respect to Dean. It is suggested that cognitive psychologists, whether working in the laboratory or attempting to do real-world studies of everyday remembering, need to avoid simplistic notions of true original events, and can do so by addressing the rhetorical organization of participants' memory accounts.