Antiepileptic medications were discontinued in 88 children with epilepsy of various causes who had been free of seizures for two to four years. The subjects were then followed for six months to five years (mean, 22 months). Sixty-six (75 per cent) remained free of seizures. Electroencephalographic characteristics, type of seizure, and age at onset were important in predicting outcome. Specific electroencephalographic features (such as the presence or absence of slowing or spikes) were more predictive than simple classification of an electroencephalogram as normal or abnormal. A history of complex partial seizures that had been controlled for two years carried a relatively favorable prognosis (P less than 0.05), whereas a history of atypical febrile seizures carried a poor prognosis (P less than 0.05). The variable of a younger age at onset was also associated with a better outcome (P less than 0.05), but only if accompanied by electroencephalographic slowing. Altogether, six variables (the interaction of age at onset with electroencephalographic slowing, electroencephalographic spikes, atypical febrile seizures, complex partial seizures, electroencephalographic slowing, and electroencephalographic change) were statistically significant (P less than 0.05) in predicting outcome. We conclude that in most children with epilepsy who have been free of seizures for two or more years, antiepileptic medications can safely be discontinued, and that it is possible to predict reasonably well which children will remain free of seizures.