The use of Bayesian algorithm on theophylline pharmacokinetics was studied in asthmatic patients. Each patient received multiple oral doses of the sustained-release theophylline tablet formulation, Theo-Dur, in an amount of 200-400 mg as maintenance dose. Plasma theophylline concentration was measured by a commercially available substrate-labeled fluorescent immunoassay. The data were analyzed by applying a one-compartment model with both zero-order and first-order absorptions based on a nonlinear least squares method, Bayesian algorithm. The present study indicated that the zero-order absorption model provided better estimated pharmacokinetic parameters than the first-order absorption model. It was also found that prediction by the Bayesian method is dependent on number of sampling trials, but the best sampling time is still unknown. This may be due to the small peak-trough plasma concentration differences. From these findings, we suggest that parameters with larger variances should be chosen by considering the interindividual and intraindividual variabilities in order to improve the accuracy of population pharmacokinetics.