Barbiturate exposure during childhood was assessed from medical records of 237 children with intracranial and spinal cord tumors and 474 matched controls in a prepaid health plan. In utero exposure was also examined in a subset of 86 "cases" and 172 controls whose mothers were health plan members during pregnancy. No association of in utero exposure to barbiturates was found [odds ratio (O.R.) = 0.96, 95% confidence interval (C.I.) = 0.47, 1.94]. An association was noted for barbiturate use during childhood (O.R. = 1.80, 95% C.I. = 1.18, 2.74) but was reduced (O.R. = 1.41, 95% C.I. = 0.89, 2.21) when history of epilepsy was taken into account and was no longer significant. An apparent dose-response effect disappeared after adjustment for a history of epilepsy. Although barbiturate use for epilepsy due to preexisting brain tumors clearly explains some of the observed association, the small, residual risk prevents us from ruling out a possible carcinogenic effect of barbiturates. Further study of cohorts of adult as well as childhood users of barbiturates and other anticonvulsants is recommended.