We assessed the prognosis for seizure control in 106 patients who were referred to an adult neurology clinic with previously untreated tonic-clonic, partial, or mixed seizures and were followed prospectively for a median of 66 months (range, 6 to 96). Twenty-six patients remained completely free of seizures for as long as they were followed. Actuarial analysis showed that 35 per cent of patients could be expected to enter a seizure-free period of at least two years at the start of treatment, 73 per cent would have had a two-year seizure-free period at the end of four years, and 82 per cent would have had a two-year seizure-free period at the end of eight years. Of 79 patients whose seizures were completely controlled for at least two years, 51 subsequently remained seizure-free. If seizures continued for up to two years after the start of treatment, the probability of subsequent seizure control fell by half. The presence of partial seizures; a high frequency of tonic-clonic seizures before treatment; a neurologic, social, or psychiatric handicap; and a family history of epilepsy each indicated a worse prognosis. We conclude that the long-term pattern of seizure control is largely established during the first two years of treatment.