A retrospective analysis of Veterans Administration (VA) Cooperative Study No. 118 was conducted to examine the relationship between serum phenytoin concentration (SPC) and various measures of patient response in 111 patients on monotherapy, mostly after 12 months of treatment. Using patient response rating scales specifically developed and validated for the VA study, we plotted SPC vs overall composite score as well as vs each of the three component ratings of neurotoxicity, systemic toxicity, and seizure type and frequency. No statistically significant association was noted between SPC and any of the response measures. We also calculated test performance characteristics for using SPC and a therapeutic range of 10-22 micrograms/ml to categorize patient status with regard to overall response, toxicity, and seizure frequency. Test sensitivities (0.13-0.53), specificities (0.62-0.81), and positive predictive values (0.08-0.44) were much lower than desirable. This results in a large proportion of false-positive and false-negative errors associated with the use of SPC as a test. Our retrospective evaluation suggests that the range of SPC values in successfully treated patients is quite broad; the value of the commonly accepted SPC therapeutic range in predicting various measures of patient response is quite limited; and patient response, therefore, should be the ultimate end point in monitoring patients on phenytoin. Prospective studies of the relationship between SPC and response should be conducted.