Investigated the extent and mechanisms of therapeutic generalization across distinct areas of agoraphobic dysfunction. Twenty-seven severe agoraphobics were each given performance-based treatment for some phobic areas while leaving their other phobias untreated. Behavioral tests revealed that (a) the treated phobias improved significantly more than the untreated (transfer) phobias, (b) the transfer phobias improved significantly more than control phobias, and (c) the transfer benefits were highly variable within and between subjects. Analyses of possible cognitive mechanisms revealed that perceived self-efficacy accurately predicted treatment and transfer effects even when alternative factors such as previous behavior, anticipated anxiety, anticipated panic, perceived danger, and subjective anxiety were held constant. In contrast, these alternative factors lost most or all predictive value when self-efficacy was held constant. Agoraphobia thus appears to be neither a unitary entity nor a mere collection of independent phobias, but a complexly patterned problem governed largely by self-perceptions of coping efficacy.