We compared a least squares regression method, used prospectively to individualise the intravenous aminophylline and oral theophylline dosage of 48 patients, with 3 other pharmacokinetic methods — Chiou’s, the steady-state clearance and the Bayesian — used retrospectively to analyse the same patient data. Methods were compared on the basis of the similarity of their parameter estimates and the accuracy with which serum concentrations during subsequent intravenous and oral therapy could be forecast, assuming each method’s parameter estimates. The least squares and Bayesian programs were able to fit data from all but 4 and 2 patients, respectively. Mean absolute prediction errors were of the order of 20% for serum concentrations during intravenous therapy, and of the order of 40% for serum concentrations during oral therapy. The accuracy of the least squares, Bayesian and steady-state clearance methods were similar, but the accuracy of Chiou’s method was comparable only when the 2 serum concentrations needed for the method were measured between 11 and 17 hours apart; an interval which corresponds to the 1.0 to 1.5 half-lives previously suggested as desirable for implementation of the Chiou method.