This study compares the academic achievement of 598 students using a 200-item multiple-choice test at the beginning of the first, third, and fifth semester in October 2005. 183 students applied for the new model curriculum with problem-based learning (PBL), of whom 103 were randomly selected (P+). The remaining 80 (P-) together with 415 students (Cl) took part in the classical curriculum based on lectures and seminars. 66 test items were related to basic sciences, 32 of these to physiology. The three groups were compared by one-way ANOVA and post-hoc analysis with Bonferroni correction. There were no differences in basic science and clinical test scores of students in the beginning of the first semester. In the third semester, P+ performed significantly better than P- and Cl in basic science (22,7±7,1 vs. 16,8±8,4 vs. 15,4±6,8; p<0,001) and clincal items (20,9±6,3 vs. 13,8±12,5 vs. 11,2±9,3; p<0,001). In the fifth semester, basic science items scored similar (27,2±6,7 vs. 28,6±6,2 vs. 26,2±7,8; p=0,52), clinical items were significantly better in P+ (29,7±9,3 vs. 19,2±8,9 vs. 16,7±9,0; p<0,001). We conclude that self-directed learning in the PBL-curriculum did not lead to weaker performances in basic science items as compared to the curriculum with systematic lectures, but was superior with respect to applied and clinical knowledge.