One thousand and thirteen patients, in remission of epilepsy for at least 2 years, were randomized to continued therapy or slow withdrawal over 6 months and were followed up for a median period of 5 years. At the time of randomization 83% of patients were receiving monotherapy with carbamazepine (237 patients), phenobarbitone/primidone (72 patients), phenytoin (184 patients) or valproate (228 patients) in low doses, and plasma levels were below the usual optimal range. The most important factor determining seizure recurrence was continued therapy, which was the case for barbiturates, phenytoin and valproate. There was no significant difference for patients taking carbamazepine at randomization, because of a low rate of recurrence in those withdrawing treatment. The difference between carbamazepine and other drugs was not explained by differences in covariate prognostic factors. There was no evidence that withdrawal of phenobarbitone was associated with withdrawal seizures. These data provide unique evidence for the effectiveness of standard antiepileptic drugs as monotherapy. The results for carbamazepine may be open to a number of interpretations.