This study investigated factors related to an initial exercise experience to explain exercise maintenance in 120 adults with rheumatoid arthritis or osteoarthritis. Integral secondary analysis was used to incorporate data from a prospective, controlled trial of exercise (Minor et al.: Arthritis Rheum 32:1396, 1989) with data collected at 18 months follow-up. The dependent variable was self-directed exercise (min/wk) reported at 3, 9, and 18 or more months after exercise class participation. Predictor variables included physical, psychosocial, disease, and programmatic factors. The all possible regressions search procedure resulted in three explanatory models (p = .0001). At 3 months the model (R2 = .45) included initial aerobic capacity, depression, and anxiety; and changes in depression and social activity. The 9-month model (R2 = .35) consisted of initial anxiety and physical activity, change in depression, support of friends for exercise, and exercise behavior at prior assessment. At 18 or more months (R2 = .42), model variables were initial aerobic capacity, change in pain, and exercise behavior at the two prior assessments. Neither disease nor program factors appeared as significant. This limited study indicates that factors associated with exercise behavior in this sample are similar to those in the general population; explanatory factors change over time, and changes ascribed to a trial behavior may influence subsequent decision making.