This article addresses a paradox. On the one hand, discourse and rhetorical studies have provided evidence that evaluative talk is both variable and rhetorically organized. On the other hand, a wide range of social psychological research is produced that both presupposes and finds evidence of enduring underlying attitudes. One explanation for this may be that, on some occasions at least, the results of attitude research are a consequence of procedures that restrict and refine from everyday evaluative practices in a way that ensures the ‘discovery’ of underlying attitudes. The article explores this explanation in one domain where there is a major practical concern with attitudes and opinions, namely market research focus groups. Detailed analysis of transcripts of eight market research focus groups identifies three procedures that moderators use to produce freestanding opinion packages: (a) they display rhetorically embedded evaluations as inconsequential; (b) they provide formal guidance for participants to produce freestanding opinions; and (c) they formulate participants' talk as freestanding opinions, stripping off rhetorical elements. The findings are supported by considering deviant cases. This illustrates one way in which evaluations are transformed into freestanding attitudes. More broadly, it contributes to a body of work that studies how social science methods work in practice.