We withdrew phenytoin from 17 inpatients maintained on combination therapy with carbamazepine for complex partial seizures and analyzed seizure occurrence in relation to plasma levels and time from initiation of withdrawal. The ratio of maximum to mean weekly seizure frequency did not vary with initial level or rate of withdrawal. The week with most frequent seizures began a median of 10 days after phenytoin levels became undetectable, and mean daily seizure frequency was higher at undetectable than at falling levels for the entire 2- to 10-week study period. Four patients had a total of 6 clusters of generalized tonic-clonic seizures; only 2 occurred while levels were falling and the other 4 at 3, 9, 28, and 42 days after reaching undetectable levels. Our data argue against the occurrence of withdrawal seizures in these patients and suggest that worsening of seizures following phenytoin discontinuation more likely reflects loss of therapeutic drug effect than a true abstinence phenomenon.