Controlled studies with new antiepileptic drugs are problematic and limited in children. Withdrawal randomization versus placebo in responders previously recognized in an open phase is a new design reported in adults which allows comparison to placebo without delaying the administration of the active compound. We applied such a design in refractory epileptic children in order to study vigabatrin (VGB) in children. Twenty-eight patients aged 1.5-18.5 years and having partially responded to VGB, prescribed in an open study for refractory epilepsy, were included. Patients were randomized to VGB (continued) or placebo (VGB blindy stopped in 3 weeks) for 2 months and seizure frequency was compared to the prerandomization period. More than 50% increase in seizure frequency induced drop-out. Fifteen patients received VGB, 13 others placebo, with the same clinical characteristics in both groups. The patients remaining in the study (primary efficacy endpoint) were more numerous on VGB (93%) than on placebo (46%) (P < 0.01) and seizure frequency (secondary endpoint) was lower on VGB than placebo (P < 0.05). The same results were observed in a subgroup of partial epilepsies. No status epilepticus was observed when withdrawing VGB and all patients returned to baseline status by reintroducing VGB. Such a randomized withdrawal design is therefore feasible in epileptic children. It provides the first VGB controlled study in this age range and demonstrates efficacy. It could be useful for future designs of drug trials in childhood epilepsy.