To compare the diagnostic performances of students in five curriculum years educated at schools with problem-based, integrative, or conventional medical curricula. Data were analyzed in 1994 for 612 students in their second, third, or fourth (preclinical) or fifth or sixth (clinical) years at three Dutch medical schools with problem-based, integrative, or conventional curricula. The students gave differential diagnoses for 30 case histories that were epidemiologically representative of Dutch society and covered all organ systems. The numbers of accurate diagnostic hypotheses were tallied for each of the groups involved. The data were analyzed using analysis of variance and post-hoc Newman-Keuls tests. Overall, the students trained within the problem-based framework and the students trained within the integrated curriculum made more accurate diagnoses than the students trained within the conventional curriculum. No overall differences were found between the students in the problem-based and integrated curricula, although the second- and third-year students from the latter performed better than the second- and third-year students from both other schools. Integration between basic and clinical sciences and an emphasis on patient problems may be the critical factors that determine superior diagnostic performance rather than whether a curriculum is self- or teacher-directed. Problem-based learning seems to live up to its expectations, but so does the integrated approach to medical education. In addition, the procedure for measuring diagnostic performance appears to be valid and to provide a simple means of measuring curriculum effects. It remains to be seen whether the findings would be replicated when students are allowed to freely gather data in open interaction with patients rather than respond to written presentations of cases.