This study examined determinants of vigorous physical activity in a community sample of adults over a 24-month period. Social learning theory predicts that causal relationships between determinants and behavior are bidirectional, and both the behavior and its influences are subject to change over time. Thus, dynamic variables were expected to account for more variance in exercise change than static baseline variables. Over 86% of respondents to the baseline mail survey completed follow-up surveys 24 months later, yielding a final sample of 1739. Two physical activity change indices were constructed that were adjusted for baseline exercise and demographic variables, so that the effects of social learning variables could be isolated. In a two-step hierarchical regression analysis, physical activity measures were regressed onto 21 static and six dynamic independent variables. Baseline self-efficacy was a significant predictor of exercise change for both dependent variables. Four dynamic social learning variables were significantly associated with both exercise change measures: self-efficacy, perceived barriers, family support, and friend support. Social learning variables accounted for 12.3-15.5% of the variance in exercise change over 24 months. The results provide strong support for bidirectional causation between behavior and determinants and suggest hypotheses to be tested in experimental studies.