We undertook a meta-analysis of the Continuing Medical Education (CME) outcome literature to examine the effect of moderator variables on physician knowledge, performance, and patient outcomes. A literature search of MEDLINE and ERIC was conducted for randomized controlled trials and experimental design studies of CME outcomes in which physicians were a major group. CME moderator variables included the types of intervention, the types and number of participants, time, and the number of intervention sessions held over time. Thirty-one studies met the eligibility criteria, generating 61 interventions. The overall sample-size weighted effect size for all 61 interventions was r = 0.28 (0.18). The analysis of CME moderator variables showed that active and mixed methods had medium effect sizes (r = 0.33 [0.33], r = 0.33 [0.26], respectively), and passive methods had a small effect size (r = 0.20 [0.16], confidence interval 0.15, 0.26). There was a positive correlation between the effect size and the length of the interventions (r = 0.33) and between multiple interventions over time (r = 0.36). There was a negative correlation between the effect size and programs that involved multiple disciplines (r = -0.18) and the number of participants (r = -0.13). The correlation between the effect size and the length of time for outcome assessment was negative (r = -0.31). The meta-analysis suggests that the effect size of CME on physician knowledge is a medium one; however, the effect size is small for physician performance and patient outcome. The examination of moderator variables shows there is a larger effect size when the interventions are interactive, use multiple methods, and are designed for a small group of physicians from a single discipline.