OBJECTIVES: To investigate the relationships among self-efficacy, changes in self-efficacy, past exercise participation, future exercise adherence, and exercise program format. METHODS: Two-year randomized trial involving subjects (n = 63) participating in an aerobic exercise program. Subjects were randomly assigned to one of three exercise conditions: higher-intensity home-based exercise, higher-intensity class-based exercise, or lower-intensity home-based exercise. RESULTS: Results indicated that baseline self-efficacy and exercise format had significant (p < .02), independent effects on adherence during the adoption and early maintenance phases of exercise behavior. In contrast, in predicting long-term exercise program maintenance, a significant (p < .05) self-efficacy x exercise format interaction indicated that self-efficacy predicted adherence only in the supervised home-based exercise conditions. Results also suggest that baseline self-efficacy, independent of the effect of past adherence, significantly (p < .03) predicted exercise adherence during the adoption phase, but not early maintenance phase, of exercise behavior. Finally, adherence change during the adoption phase of exercise behavior significantly (p < .04) predicted Year-one levels of self-efficacy even after adjusting for the effect of baseline self-efficacy. CONCLUSIONS: These results suggest that exercise program format as well as an individual's initial cognitive and behavioral experiences in an exercise program play significant roles in determining exercise adherence.