Two different doses of sodium valproate (VPA), 500 mg b.i.d. and 1,000 mg b.i.d. as enteric-coated tablets, were used in this randomized, double-blind, cross-over monotherapy study of 16 patients with juvenile myoclonic epilepsy. Observation time was 6 months on each dose and included admittance for a 12-h serum concentration-time curve. There was a nonlinear relation between dose and concentration, with a negative deviation from the linear relation for total concentration and a positive deviation for the unbound fraction. Clearance for total concentration increased during high-dose treatment, but intrinsic clearance did not differ between doses. We measured the variation of repeated total and unbound VPA concentrations in up to 6 monthly samples on each dose. The coefficient of variation was 20.7% for total and 29.9% for unbound concentration on the lower dose, and 16.5% for total and 28.5% for unbound concentration on the higher dose. This difference between doses is not statistically significant. There was good correlation between the concentration taken before morning dose and AUC for one dose interval, especially during high dose, but the morning concentration was not the trough level. We conclude that the pharmacokinetic requirements for therapeutic drug monitoring of VPA are established.