Summary Until now, no evaluation of phenytoin dosing methods has been undertaken in a large group of patients, to our knowledge. The goal of this study was to determine which of eight different dosing methods most accurately predicts a phenytoin steady-state concentration. Seventy-six patients were chosen, retrospectively, from a state-funded institution for the mentally retarded. Eligibility criteria included two or more different doses of phenytoin and corresponding plasma concentrations. Relative predictive performance was determined by comparing results of simple linear regression. Also, relative bias and precision were determined by comparing mean prediction errors, root mean squared errors, and respective 95% confidence intervals. Of the methods requiring one dose-concentration pair, Rambeck's nomogram was the best predictor of phenytoin concentrations. The methods requiring two known doses and plasma concentrations were more accurate. Their predictive performance was equivalent, although use of the Tozer equation might be preferred for its convenience. None of the methods tested were sufficiently precise to substitute for confirmatory serum phenytoin concentrations.