The population pharmacokinetics of theophylline were studied in 55 patients with stable chronic airway obstruction. Two hundred and seventy six theophylline serum concentrations after intravenous short infusion were analyzed using a nonlinear mixed-effect model. The influence of hepatic dysfunction, smoking habit, age and the measurement of arterial blood gases (oxygen tension : PaO₂, carbon dioxide tension : PaCO₂, blood pH) and clinical laboratory tests (serum albumin concentration, haematocrit) on the pharmacokinetic parameters of theophylline was examined by the likelihood ratio test. Assessment of each factor was made by a forward selection method. In the final regression model, the total body clearance (CL, l/h/kg) was related to the value of PaCO₂ as well as to the presence of hepatic dysfunction, and the volume of distribution (V_d, l/kg) was related with the PaCO₂ value as expressed in the following equations : CL=exp(-3.78-0.525·HF+0.0233·PaCO₂) and V_d=exp(-1.12+0.00934·PaCO₂), where HF is a categorical variable with a value of unity if a patient has hepatic dysfunction otherwise zero. The interactions among blood gas measurements were observed and the CL and V_d of theophylline would be inversely correlated with PaO₂ or pH, if we selected PaO₂ or blood pH to be a more important factor than PaCO₂. The inter-individual variabilities in CL and V_d were 38.5% and 12.5%, respectively, and the residual variability in theophylline serum concentrations was 10.6% as a coefficient of variation. This final model and the population parameters of theophylline will be useful for individualization of a drug dosage regimen by means of the Bayesian method. Significant correlation between arterial blood gas measurements and pharmacokinetic parameters observed in this study indicates that the pharmacokinetics of theophylline can vary in association with the severity of respiratory diseases.