BACKGROUND: Students' learning was used as an outcome measure in the first phases of the major curriculum reform started in 1987 by the Université de Sherbrooke Faculty of Medicine, which shifted from a traditional to a student-centered, problem-based learning (PBL) and community-oriented program. The system for evaluating preclinical students' learning is intended to reinforce the integration of basic and clinical sciences. METHOD: To discover whether the evaluation system was fulfilling its intended goals, the authors used data from the classes of 1991-1993 to assess the reliability and validity of three evaluation instruments. The three instruments were (1) written examinations composed of multiple-choice questions (MCQs), short-answer questions (SAQs), and problem-analysis questions (PAQs); (2) PBL tutor rating forms that evaluate students' reasoning skills, communication and group-interaction skills, and autonomy and humanism; and (3) clinical skills evaluations, including objective structured clinical examinations (OSCEs). The weights allocated to the instruments reflected how the faculty valued each evaluation dimension in each of the three phases of the preclinical curriculum. RESULTS: Reliability indexes improved throughout the system implementation. The written examinations proved to have content validity according to the PBL learning objectives. As evaluated by students, the PAQs were found to be at a taxonomic level that assessed ability to analyze information a third of the time in the first year of implementation of the PBL curriculum and 17% in the second year. Variations and correlations of students' mean performances across instructional units and between the evaluation instruments led to the development of a student longitudinal performance profile to be used before yearly promotion decisions are proposed. The profile was introduced in the fifth year of PBL implementation. CONCLUSION: The system allows students to learn higher-taxonomic-level content and fulfills the institution's social responsibility of judging program outcomes and promoting qualified students, although evaluation by PBL tutors is still psychometrically questionable and the measurement of students' reasoning and ability to analyze problems is still an unfinished evaluation task