Moderate and high-dose methotrexate was given by a 4-h infusion to 10 patients. The pharmacokinetics of methotrexate, determined by fluorescence polarization immuno-assay, was successively described by a 2- and a 3-compartment open model with elimination from the central compartment. Population pharmacokinetic parameters were obtained by non-linear regression from 8 data points for each course and were subsequently used to fit the same data by the Bayesian estimation method. According to Akaike's information criterion, the 3-compartment model was found statistically superior in 7 patients out of 10. Using this model clearance was well predicted (+/- 5%) by the Bayesian method with only 2 points taken at the end of the infusion and 24 h after. The prediction was less good for the steady-state volume of distribution (+/- 18%) and the half-lives (+/- 20-30%). This procedure enables a good estimation of individual pharmacokinetic parameters for methotrexate, specially with clearance, at minimal cost and minimal disturbance for the patient.