A prospective multicentre randomised study of continued antiepileptic treatment vs slow withdrawal was conducted in 1013 patients who had been free of seizures for at least 2 years. Comparison of randomised and eligible, but non-randomised, patients suggests the results should be applicable to a wider patient population. By 2 years after randomisation, 78% of patients in whom treatment was continued and 59% of those in whom it was withdrawn remained seizure free, but thereafter the differences between the two groups diminished. Non-compliance with continued treatment accounted for only a small proportion of the risk to the group continuing with treatment. The most important factors determining outcome were longer seizure-free periods (reducing the risk) and more than one antiepileptic drug and a history of tonic-clonic seizures (increasing the risk). Other factors (eg, history of neonatal seizures, specific electroencephalographic features) seemed to have smaller effects, but even in such a large study the confidence intervals for these observations were wide.