A prospective study tested the stability of the IQ in children with seizure disorders. Seventy-two children with epilepsy underwent psychological evaluations within two weeks of initial diagnosis and yearly thereafter for an average of 4 years. Forty-five of the patients also had a nonepileptic sib evaluated in parallel. The mean IQ for all the children with epilepsy was 99.7 (±20.2, standard deviation) at the time of the initial test, not significantly different from the siblings. This score did not change appreciably with time. Eight of the 72 epileptic patients (11.1%), however, had a persistent decrease in IQ of 10 points or more. These patients had a higher incidence of drug levels in the toxic range (p <0.001), their epilepsy was more difficult to control (p <0.005), and their seizures began at an earlier age (p <0.05). Discriminant analysis revealed that the number of drugs to which the patient became toxic and the age at seizure onset were the two best predictors of ultimate IQ. These two predictors correctly classified 71% of all patients as to whether their IQ would drop by 10 or more points during the test period, remain within 10 points of the initial test score, or increase by more than 10 points. Total number of seizures and seizure control were less good predictors, according to this method of analysis. The findings suggest that, in younger children in particular, total seizure control should not be achieved at the price of repeated episodes of drug toxicity.