Several authors have argued that memory-based reports about dispositional characteristics of people exhibit a correlational structure that is unrelated to the co-occurrences of pertinent on-line recorded behaviors. We hypothesize, however, that the empirical evidence, deemed to be in favor of this challenging hypothesis, is due to the neglect of act overlap among the behavior classes under investigation. The present study was conducted in order to examine this hypothesis. Eight groups, each composed of 6 male actors, were videotaped while discussing controversial topics. The tapes were later shown to five judges who retrospectively estimated, with respect to 16 behavior categories, the individual act frequencies after viewing each discussion in its entirety. Four other judges classified each activity, immediately after observing it, by using one of two coding schemes. Two judges made forced choices; that is, they assigned a given behavior to exactly one category. Two other judges rated the prototypicalities of each activity with respect to each of the 16 behavior categories. The latter coding scheme led to substantially higher correspondences between the correlational structures of the retrospectively estimated and the on-line-recorded act frequencies. The results support a systematic overlap hypothesis rather than a systematic distortion hypothesis.