Marketing research surveys often elicit behavioral frequency reports. When estimating the number of times a respondent engages in a behavior, s/he may use information about the behavior stored in memory, information provided by the response context, or both. Based on an accessibility-diagnosticity framework, we theorize that the probability of using context-based information in forming a frequency judgment is inversely proportional to the diagnosticity of the alternate inputs accessible in memory. That is, when memory-based information is accessible and diagnostic, contextual information is not used; when memory-based information is accessible but not diagnostic, the use of contextual information depends on its perceived diagnosticity. Finally, when memory-based information is not accessible, contextual information is used even when its diagnosticity is questionable. The results of three experiments support this model. Theoretical implications and recommendations for questionnaire design are discussed.