The purpose of this review is to synthesize all available evaluative research from 1970 through 1992 that compares problem-based learning (PBL) with more traditional methods of medical education. Five separate meta-analyses were performed on 35 studies representing 19 institutions. For 22 of the studies (representing 14 institutions), both effect-size and supplementary vote-count analyses could be performed; otherwise, only supplementary analyses were performed. PBL was found to be significantly superior with respect to students' program evaluations (i.e., students' attitudes and opinions about their programs)--dw (standardized differences between means, weighted by sample size) = +.55, CI.95 = +.40 to +.70 - and measures of students' clinical performance (dw = +.28, CI.95 = +.16 to +.40). PBL and traditional methods did not differ on miscellaneous tests of factual knowledge (dw = -.09, CI.95 = +.06 to -.24) and tests of clinical knowledge (dw = +.08, CI.95 = -.05 to +.21). Traditional students performed significantly better than their PBL counterparts on the National Board of Medical Examiners Part I examination--NBME I (dw = -.18, CI.95 = -.10 to -.26). However, the NBME I data displayed significant overall heterogeneity (Qt = 192.23, p < .001) and significant differences among programs (Qb = 59.09, p < .001), which casts doubt on the generality of the findings across programs. The comparative value of PBL is also supported by data on outcomes that have been studied less frequently, i.e., faculty attitudes, student mood, class attendance, academic process variables, and measures of humanism. In conclusion, the results generally support the superiority of the PBL approach over more traditional methods. Acad. Med. 68 (1993):550-563.