SUMMARYThe major focus in the selection of entrants for medical school has traditionally been on academic achievement in school‐leaving examinations in which certain science subjects are a requirement. A longitudinal study of 413 successful applicants was undertaken to determine the relationship of these admission criteria to subsequent performance. The findings supports a correlation between overall marks in the school‐leaving examination and the annual Grade Point Averages. Those students in the top quartile for marks showed a significant advantage in terms of achievement but only in the preclinical years. Despite the significant correlations no predictions could be made on the basis of overall marks. No correlation was found with levels of clinical competence during the ward clerkships or with the interdisciplinary objective structured clinical examination (OSCE) in the final examination. Marks in individual school‐leaving examination subjects correlated with performance during different parts of the course but only those entrants in the top quartile for marks in physics and biology showed an advantage through to the clinical years. English marks were the least correlated and failed to confer an advantage in any year of the course. None of the correlations between school‐leaving marks and grades in medical school exceeded 0.4. The predictive value of school‐leaving examination marks therefore accounted for only 16% of the variance in subsequent examinations. Selection of medical students on the basis of academic criteria alone is inadequate and should be accompanied by assessment of personal qualities. This School no longer uses school‐leaving marks as the primary selection instrument. Measurement of personal qualities, motivation and life experiences by structured interviews is combined with intellectual ability to rank applicants. Correlational or predictive studies cannot of themselves decide which individual subjects should be prerequisites. However, the failure of any one subject to demonstrate greater correlation or prediction tends to support the study of a broader range of subjects by potential