A total of 20 children with various types of epilepsy were treated with valproate, 11 with monotherapy and 9 with valproate in combination with phenobarbitone, phenytoin, or carbamazepine. Valproate was given either every 8 or 12 h. At least two different dose levels were tried in each patient. The pharmacokinetics of valproate during the interval between doses was determined using a gas chromatographic technique. The clinical effect of the treatment was assessed by interviewing the parents. The plasma concentrations showed considerable fluctuation during the intervals between doses. The mean increase from pre-administration to peak level was 82% when the dose interval was 12 h, and 62% when it was 8 h. The mean plasma half-life of valproate, using a one-compartment model, was 10.9 +/- 1.3 h (mean +/- SD). The plasma half-life of valproate was decreased when the drug was combined with the other anti-epileptics. The calculated area under the concentration versus time curve was linearly related to dose, both in a single patient on four dose levels and when different patients were compared. The clinical effect of valproate monotherapy was best in patients with absences, usually good in myoclonus and less favourable in other types of epilepsy. For children with absences, the optimal dose range of valproate was between 20 and 40 mg/kg/24 h. In comparison, the myoclonic types of epilepsy needed a slightly higher dose level, between 30 and 60 mg/kg/24 h. In the latter group a "therapeutic window" seems to exist, since patients below and above the suggested dose levels were not well-controlled. Therapeutic monitoring of valproate does not appear meaningful when the drug is used as monotherapy. However, in combination therapy, determination of the plasma levels of all anti-convulsants used may be helpful. The large fluctuations of valproate during a dose interval must be taken into consideration when the clinical effects are analysed.