Although the long hours worked by resident physicians have raised concern in the public, legislators, and members of the medical profession, the consequences of sleep loss on the ability of residents to perform in clinical settings is unknown. The present study examined the effects of various amounts of reported sleep on cognitive performance measured by the American Board of Family Practice in-training examination. A total of 353 family practice residents in 21 programs who took the examination in Pennsylvania in 1988 were studied. Linear regression analysis demonstrated a statistically significant (P less than .05) decline in composite test score with decreasing sleep on the night before the examination for residents in each year of training. Loss of one night's sleep resulted in changes in test scores that were approximately equivalent in magnitude to the change that occurred in test scores between residents in the first and third year of training. The results suggest that prolonged testing over several hours may be necessary to detect the subtle but significant differences in cognitive performance that are present with relatively mild degrees of sleep loss.