Clinical effects at three different serum levels of sodium valproate (VPA) were compared in a triple-blind, multiple crossover trial comprising 13 epileptic inpatients. Patients were selected regardless of seizure type, and all were in concomitant antiepileptic treatment, which was kept constant throughout the study. A significant relationship between the decrease in number of seizures and increasing VPA serum level was demonstrated. The relationship between VPA dose and serum level was curvilinear. Statistical evaluation of patients by seizure type in relation to clinical effect of VPA was only possible for secondary generalized seizures. Between phenytoin, phenobarbital, and carbamazepine and the different VPA serum levels no interactions could be demonstrated. Recorded side effects were always mild and transient. No obvious correlation between side effects and VPA serum level was established.