1. Phenytoin protein binding in epileptic patients on phenytoin as monotherapy has been compared with protein binding in patients treated with both phenytoin and sodium valproate. In addition the relative value of assayed total phenytoin plasma concentrations and assayed unbound phenytoin plasma concentrations and the value of predicted unbound phenytoin plasma concentrations in predicting phenytoin toxicity has been assessed. 2. The mean phenytoin unbound fraction for patients taking sodium valproate (0.122) was significantly greater than for those on monotherapy (0.082). 3. There were six episodes of clinical toxicity. In five toxic episodes the assayed unbound phenytoin plasma concentration was a better reflection of toxicity than the assayed total phenytoin plasma concentration, and four of these occurred in patients on sodium valproate. 4. Unbound phenytoin plasma concentrations were predicted from a single regression equation correlating all assayed total phenytoin plasma concentrations with assayed unbound phenytoin plasma concentrations, from two separate regression equations for each group of patients, and from the correlation between phenytoin protein binding and plasma albumin concentration. 5. The unbound phenytoin plasma concentrations predicted from the two regression equations were statistically no less effective than the assayed unbound phenytoin plasma concentrations in assessing toxicity. 6. Despite a correlation between plasma albumin concentrations and phenytoin protein binding, the use of albumin concentrations in predicting unbound phenytoin plasma concentrations appeared to be of little additional benefit.