OBJECTIVE: To compare clinical and demographic characteristics of individuals self-selecting yoga or physical therapy (PT) for treatment of chronic low back pain (cLBP) and to examine predictors of short-term pain and functional outcomes. DESIGN: Descriptive, longitudinal study. SETTINGS: A hospital-based clinic that offers modified integral yoga classes for cLBP and 2 outpatient PT clinics that offer exercise-based PT. PARTICIPANTS: Adults (n=53) with cLBP?12 weeks: yoga (n=27), PT (n=26). METHODS: Yoga participants attended a 6-week, once weekly, 2-hour yoga class. PT participants underwent twice weekly, 1-hour individualized PT. Data were collected at baseline and at 6 weeks. Groups were compared by using ?2 and independent samples t-tests. Hierarchical linear regression was used to predict treatment outcomes. MAIN OUTCOME MEASURES: Disability (Roland Morris Disability Questionnaire), health status (Rand Short Form 36 Health Survey 1.0), pain bothersomeness (numerical rating scale), back pain self-efficacy (Back Pain Self-Efficacy Scale), and treatment satisfaction. RESULTS: At baseline, yoga participants were significantly less disabled (P=.013), had higher health status (P=.023), greater pain self-efficacy (P=.012), and less average pain bothersomeness (P=.001) compared with PT participants. At 6 weeks, when controlling for baseline group differences, greater pain self-efficacy was the strongest predictor for reduced pain and higher function for the entire sample. A significant group interaction by baseline pain self-efficacy predicted disability at 6 weeks. PT participants with low pain self-efficacy reported significantly greater disability than those with high pain self-efficacy. Yoga participants with low and high pain self-efficacy had similar disability outcomes. CONCLUSION: These findings strengthen evidence that self-efficacy is associated with cLBP outcomes, especially in individuals self-selecting PT. Further research to evaluate outcomes after yoga and PT in participants with low pain self-efficacy is needed.