Previous research suggests that perceivers estimate the cause of behavior by (a) identifying the behavior, (b) inferring the actor's disposition from the behavior, and (c) correcting this inference for situational constraints. An experiment investigated whether perceivers' inferential goals alter this process. Perceivers viewed a silent videotape of an anxious interviewee. Some perceivers estimated the target's dispositional anxiety; others estimated the degree of anxiety provoked by the interview questions. Within these conditions, half simultaneously performed a cognitive rehearsal task. Of perceivers who estimated the target's dispositional anxiety, those who performed the rehearsal task inferred more dispositional anxiety. In contrast, of perceivers who estimated the anxiety provoked by the questions, those who performed the rehearsal task inferred less dispositional anxiety. These findings suggest that social inference is more flexible than previously thought.