Although there are data on the duration of seizures in patients with refractory epilepsy, little is known about the duration of seizures in nonrefractory epilepsy populations. In a prospective study, seizure duration was determined in 407 children with a first unprovoked seizure using a structured interview and review of medical and ambulance records. Analysis focused on the distribution of seizure duration and on the conditional probability that a seizure would stop once it had already lasted for a specified time. Seizures lasted > or = 5 minutes in 50% of cases, > or = 10 minutes in 29%, > or = 20 minutes in 16%, and > or = 30 minutes in 12%. Seizure duration data were best fit as the sum of two exponential distributions, one with a mean of 3.6 minutes accounting for 76% of cases and the other with a mean of 31 minutes accounting for 24% of cases. The longer a seizure lasted, the less likely it was to stop within the next few minutes. In the 182 children with 2 or more seizures, the durations of the first and second seizures were highly correlated (r = 0.395, p < 0.0001). We conclude that the distribution of seizure duration in children with a first unprovoked seizure differs markedly from that observed in patients with refractory epilepsy. A subgroup of patients are predisposed to prolonged seizures. The data suggest that, once a seizure lasts for more than 5-10 minutes, it is unlikely to stop spontaneously within the next few minutes, and intervention is therefore indicated. These findings also support the continued use of the current definition of status epilepticus as a seizure lasting for 30 minutes or longer for epidemiologic studies.