The central tenet of social cognitive theory, that individuals' construal processes and behaviors mutually reinforce each other, is tested for exercise behavior. Two longitudinal data sets (year 1 to year 2 and year 1 to year 6) from the Stanford Five-City Project, a field experiment to promote cardiovascular disease prevention in California, are analyzed through structural equation modeling techniques to evaluate the effects of demographics, exercise knowledge, and exercise self-efficacy on exercise behavior. The effects of exercise behavior on subsequent knowledge and self-efficacy are also examined. In both data sets (year 1 to year 2, N = 1254 and year 1 to year 6, N = 939), education, income, age, and sex were significant predictors of exercise behavior, Self-efficacy and knowledge also predicted behavior. Prior exercise behavior predicted subsequent knowledge and self-efficacy. Prior knowledge and self-efficacy, in turn, predicted subsequent exercise behavior. Recommendations are made for enhancing the effectiveness of public health efforts designed to promote healthy behaviors.