Problem-based learning curricula are growing in popularity, and questions have been raised about the appropriateness of standardized examinations, such as the National Board of Medical Examiners (NBME) Parts I, II, and III examinations, for assessing students in these new curricula. Data on students' performances on the NBME I were analyzed for 508 graduates of the conventional track and 167 graduates of the problem-based Primary Care Curriculum (PCC) track at the University of New Mexico School of Medicine from the classes of 1983-1992; on NBME II, for 447 and 144 graduates, respectively (classes of 1983-1991); and on NBME III, for 313 and 100 graduates, respectively (classes of 1983-1989). The analyses also included data on the students' total Medical College Admission Test (MCAT) scores, undergraduate science grade-point averages (SGPAs), and admission subgroups within tracks. The statistical methods included analysis of covariance, Student's t-test, and the Fisher exact test. The students who had requested the PCC track but had been randomized into the conventional track had the highest mean scores on all the study variables (for 34 students, 521 on the NBME I, and for 19 students, 551 on the NMBE III). The high-risk students who had requested but had not been accepted into the PCC track seemed to benefit from the highly structured conventional track with regard to their NMBE I performances (467 for 18 students). The PCC students--both those who had been randomized into the PCC and those who had been selected into the PCC--had significantly lower mean scores on the NBME I (455 for 85 students and 463 for 82 students compared with 505 for the 439 students who chose the conventional track), but significantly higher mean scores on the NBME III (521 for 38 students and 522 for 62 students compared with 483 for the seven high-risk students and 487 for the 276 students who chose the conventional track). For both tracks, strong relationships were found among the scores on the three NBME examinations. For the PCC students, significantly weaker relationships were found between mean SGPAs and mean scores on the NBME I, II, and III. For both tracks, MCAT scores, especially in the lowest and highest ranges, were most predictive of performances on the NBME I and II. In the short run, the more teacher-centered and structured conventional curriculum better prepared the students for the NBME I, while in the long run, the more student-centered problem-based curriculum better prepared the students for the NBME III: