ABSTRACT This study investigates how frequency estimates of behaviors and trait-related activities, trait ratings on unipolar scales, and trait ratings on bipolar scales reflect behavior observations. Four discussion groups, each with six participants, were videotaped while discussing controversial topics. On these tapes, 2,149 activities were identified, and the prototypicality of the activities for 16 behavior-descriptive verbs and 8 trait-descriptive adjectives was estimated. The prototypicality ratings were used to identify each disputant's frequency of discontinuing, unrelated, and confirming instances. These frequencies were then correlated with (a) act frequency estimates for the behavior categories (e.g., “X agreed … times”), (b) act frequency estimates for adverbially described activities (e.g., “X acted cooperatively … times”), (c) trait ratings on unipolar scales (e.g., “How cooperative is X?”), and (d) trait ratings on bipolar scales (e.g., cooperative-antagonistic). All global ratings except for the trait ratings on bipolar scales reflected differences in overall activity to some extent. With overall activity partialled out, however, more disconfirming and unrelated activities caused lower act frequency and trait ratings, whereas more confirming instances caused higher ratings.