The purpose of this study was to test the (mis)match model of Rachman and co-workers during real life exposure therapy in panic disorder patients with agoraphobic avoidance. The results showed that although the patients tended to overpredict their expected fear before the exposure sessions, their predictions did not show a tendency to become more accurate in the course of treatment. After an overprediction predicted fear tended to decrease and self-efficacy tended to increase; after a correct match both predicted fear and self-efficacy remained unchanged. Patients reports of fear tended to decrease within and across exposure sessions regardless of the occurrence of correct or incorrect matches. Patients reported significantly more positive self-statements in underprediction sessions compared to overprediction and correct match sessions. Positive self-statements increased both in underprediction and correct match sessions, but decreased in overprediction sessions. The largest reduction in negative self-statement within an exposure session was found in correct match sessions. The theoretical and clinical relevance of the results are discussed in line of the match/mismatch model of Rachman and co-workers.