Even with the passage of state safety belt laws, primary care physicians can contribute to their patients' safety by brief interventions. The present study explores the prevalence of such action with adult patients and tests the power of constructs taken from social learning theory to explain physicians' behavior. These constructs included self-efficacy, personal behavior (self-modeling) and three outcome expectations--expectation of patient follow-through, health impact, and impact of health promotion on the practice. Data were taken from a survey of Texas family physicians prior to enactment of the state law (n = 209). History-taking and advising were combined to form a single scale, "safety belt action." Prevalence of safety belt action was low. Overall, only 5% said they ask routinely about safety belts; 58.1% do not advise or discuss the risk even when they are aware of nonuse. Social learning theory variables accounted for 34% of the variance in safety-belt action after controlling for year of graduation in a hierarchical regression analysis. Self-efficacy was entered first, and it predicted 25% of the variance. The other social learning variables were entered together, and they predicted the additional 9% of the variance after controlling for year of graduation and self-efficacy. Of these other variables, only health impact was significant, however. These findings suggest several avenues for improving safety belt action and add evidence for the importance of outcome expectations over and above self-efficacy.