Levels of morbidity in 789 children 7 to 11 years of age attending two primary care pediatric clinics in a health maintenance organization were examined in relation to psychiatric disturbance. Physical morbidity was measured as mean number of illness episodes per year enrolled, based on the child's medical record. Two measures of psychiatric disturbance were compared: the pediatricians' judgment and a detailed assessment using standard psychiatric interviews with parent and child. Children identified by pediatricians as disturbed had more than twice as many physical illness episodes as nonidentified children. Children identified by the standard psychiatric assessment had the same number of physical illness episodes as nondisturbed children. Pediatricians showed high specificity but low sensitivity to mental illness. Their sensitivity in the high user group was double that in the low user group. These results suggest that (1) the association between mental illness and high use may be, in part, the result of the confounding factor of physicians' judgment; (2) in settings where primary care practitioners serve as "gatekeepers" to mental health services, the offset effect of lower medical service use following psychiatric treatment may be partially explained by this; (3) the source of referral must be taken into account when assessing the offset effect in other settings.