This study assessed the ability of a computer program Simulated Kinetics (SIMKIN) to predict serum theophylline concentrations in ambulatory patients receiving oral theophylline. Data were collected by retrospective review of prospectively obtained data. A total of only 20 measured serum theophylline concentrations could be included in the study, although records of 195 patients were reviewed. An estimated patient compliance of 90-110% was required and was computed using prescription refill information. Predicted serum theophylline concentrations were generated for each patient by entering into the SIMKIN program the characteristics pertinent to theophylline disposition and the patient's theophylline dosing regimen. Actual and SIMKIN-predicted theophylline concentrations were compared by using simple linear regression and by constructing a 95% confidence interval around the mean prediction error and root mean squared error. The ability of SIMKIN to predict therapeutic category, i.e., subtherapeutic, therapeutic, or toxic, was assessed using Fisher's exact test. SIMKIN predictions of individual theophylline concentration were insufficiently accurate to replace confirmatory followup monitoring of actual levels. However, SIMKIN was able to predict the therapeutic category with 70% accuracy. We conclude that SIMKIN may be useful for categorizing a dose regimen of theophylline as therapeutic, but that it is of little use in predicting individual concentrations in outpatients when literature-averaged pharmacokinetic parameters are the sole criteria for prediction, and compliance cannot be accurately assessed.