The purpose of the study was to compare the impact of continuing medical education for primary care physicians in Halifax, Nova Scotia, Canada, delivered through a problem-based learning (PBL) format with that of a lecture-based format, in the clinical area of headache diagnosis and management. From January to June 1995, 38 physicians participated in three problem-based learning sessions, and 49 in a lecture-based session in the clinical area of headache. Pre- and post-test assessments of knowledge were made of each group before and immediately after the educational sessions. A second evaluation using Key Features Problems (KFP) to measure clinical reasoning was administered to both groups 3 months later. Analysis of covariance between groups on the post-test, using the pre-test result as covariate, showed significantly greater knowledge in the PBL group. Mean post-test KFP scores also were significantly higher for the PBL group. The PBL group's satisfaction with several programme dimensions was significantly higher than that of the lecture group. These results must be viewed in the context of differing amounts of exposure for the two groups. Problem-based learning in continuing medical education in the area of headache management was associated with greater knowledge acquisition and with greater improvement in clinical reasoning skills than in a lecture-based approach. Problem-based learning also was preferred by family physicians. However, the cost-benefit of this approach was questioned, since the PBL group had more exposure. Several factors limit the generalizability of this study, e.g. the impossibility of randomly composing the two experimental groups. Further research is required to determine if the results are generalizable and whether a PBL approach results in change in practice behaviour.