The utility of predicting theophylline clearance (CL) from two serum concentrations obtained during continuous intravenous aminophylline infusion was examined in 16 stable, adult patients. Blood for theophylline measurement was obtained 0, 6, and 12 h after starting infusions and, thereafter, at 12-h intervals. EMIT was used to assay samples in multiple runs as they were obtained. Later, each sample was reassayed by EMIT within a single run. Bayesian least-squares regression and the algebraic method of Chiou were used to predict CL using the 0,6 and 0,12 h concentrations. "Actual" CL was measured by nonlinear least-squares regression of all concentrations obtained during prolonged infusions. Prediction bias and precision were assessed by calculating mean percent error (PCE) and mean absolute percent error (APCE), respectively. A three-way repeated-measures ANOVA was used to examine the effect of the method of CL prediction, assay procedure, and time interval between samples on PCE and APCE. Bayesian predictions were less biased and slightly more precise than Chiou predictions. The assay procedure had no effect on bias but precision was improved using a single-assay run. Predictions were less biased and more precise with 0,12 h versus 0,6 h data. Serum samples for theophylline measurement should be obtained after initiating constant intravenous aminophylline and again 8-12 h later in stable, adult patients. Prediction of CL with either of the concentration-based methods studied will then allow safe and rapid adjustment of dosage to achieve therapeutic serum concentrations.