We studied the effects of valproate (VPA) on local cerebral glucose metabolism (LCMRglc) in eight patients with partial seizure disorders and two with primary generalized epilepsy. Each patient had two positron-emission tomography (PET) scans with 18F-2-deoxyglucose (FDG), with, and without, VPA (mean level 52 mg/dl, range 30-127 mg/dl). Patients continued carbamazepine (CBZ) for both scans: serum concentrations were not significantly changed by VPA (CBZ range 5.4-12 mg/dl). Seven patients had the "without-VPA" scan first. Mean interval between PET scans was 75 days. Global CMRglc was decreased by 22% by addition of VPA (7.2 +/- 1.8 mg/100 g/min without VPA, 5.6 +/- 1.1 g/min with VPA, p less than 0.05, corrected). Thirteen regions of interest (ROIs) were analyzed in each hemisphere in each PET scan. Metabolic rates were significantly lower in 15 of 26 ROIs with VPA (p less than 0.05, corrected). VPA depresses cerebral metabolism to a greater degree than do CBZ and phenytoin (PHT) but less than does phenobarbital (PB). The metabolic effect may be related to the mechanism of action and have neuropsychological implications.