When asked to report on behaviors and experiences, participants draw on the specified reference period to infer the question meaning: Short reference periods suggest that the question pertains to frequent experiences; long reference periods suggest that it pertains to rare ones. Because frequent experiences are typically less intense than rare ones, this meaning shift results in reports of different experiences. Three experiments support this analysis in the domain of emotion reports. Participants asked how frequently they get angry (a) assumed that the question refers to less intense and more frequent episodes when presented with a short (1-week) rather than a long (1-year) reference period, (b) reported more extreme episodes in the latter case, and (c) provided differential frequency reports. These differences reflect conversational inference processes and cannot be fully accounted for by memory search biases.