Social learning theory has generated two different approaches for the assessment of expectancies. Bandura argues that expectancies are specific and do not generalize. Therefore, he prefers measures of specific efficacy expectations. Others endorse the role of generalized expectancies measured by locus of control scales. The present study examines specific versus generalized expectancies as mediators of changes in exercise behavior among 60 older adult patients with Chronic Obstructive Pulmonary Disease. The patients were given a prescription to increase exercise and randomly assigned to experimental groups or control groups. All groups received attention but only experimental groups received training to increase their exercise. After 3 months, groups given specific training for compliance with walking significantly increased their activity in comparison to the control group receiving only attention. These changes were mediated by changes in perceived efficacy for walking, with efficacy expectations for other behaviors changing as a function of their similarity to walking. A generalized health locus of control expectancy measure was less clearly associated with behavior change. The results are interpreted as supporting Bandura's version of social theory.