Summary: Little is known about what factors predict intractable epilepsy at the time of initial diagnosis. We performed a case-control study to identify early predictors of medically intractable epilepsy in children. Cases were children who had an average of one seizure or more a month over a 2-year period and who, during that time, had failed trials of at least three different antiepileptic drugs (AEDs). Controls were children who had epilepsy, who had been seizure-free for 2 years, and who had never, before becoming seizure-free, met the definition for intractable epilepsy. Strong univariate associations were noted between intractability and several factors: infantile spasms (IS) remote symptomatic epilepsy, a history of status epilepticus (SE) before the diagnosis of epilepsy, neonatal seizures, and microcephaly. Cases were significantly younger than controls at onset (1.8 vs. 5.8 years); this was not due solely to cases with onset during the first year of life but was an association apparent throughout the age range studied. With multiple logistic regression, independent predictors of intractability were IS, odds ratio (OR) = 10.42, p = 0.03; age at onset with a decreasing risk with increasing age, OR = 0.77 per year, p < 0.0001; remote symptomatic epilepsy, OR = 2.24, p = 0.04; and SE, OR = 3.30, p = 0.04. These findings complement those of recent cohort studies of remission of epilepsy and provide useful leads for future prospective studies of intractable epilepsy.