Given the continuous advances in the biomedical sciences, health care professionals need to develop the skills necessary for life-long learning. Self-directed learning (SDL) is suggested as the methodology of choice in this context. The purpose of this systematic review is to determine the effectiveness of SDL in improving learning outcomes in health professionals. We searched MEDLINE, EMBASE, ERIC and PsycINFO through to August 2009. Eligible studies were comparative and evaluated the effect of SDL interventions on learning outcomes in the domains of knowledge, skills and attitudes. Two reviewers working independently selected studies and extracted data. Standardised mean difference (SMD) and 95% confidence intervals (95% CIs) were estimated from each study and pooled using random-effects meta-analysis. The final analysis included 59 studies that enrolled 8011 learners. Twenty-five studies (42%) were randomised. The overall methodological quality of the studies was moderate. Compared with traditional teaching methods, SDL was associated with a moderate increase in the knowledge domain (SMD 0.45, 95% CI 0.23-0.67), a trivial and non-statistically significant increase in the skills domain (SMD 0.05, 95% CI-0.05 to 0.22), and a non-significant increase in the attitudes domain (SMD 0.39, 95% CI-0.03 to 0.81). Heterogeneity was significant in all analyses. When learners were involved in choosing learning resources, SDL was more effective. Advanced learners seemed to benefit more from SDL. Moderate quality evidence suggests that SDL in health professions education is associated with moderate improvement in the knowledge domain compared with traditional teaching methods and may be as effective in the skills and attitudes domains. © Blackwell Publishing Ltd 2010.