The predictive performance of a two-compartment Bayesian forecasting method for lidocaine (L) was evaluated concurrently with lidocaine therapy in 46 hospitalized patients; 14 of these patients presented with congestive heart failure (CHF). Using an HP-85 microcomputer, demographic and dose-concentration information obtained during continuous lidocaine therapy was used to forecast subsequent lidocaine concentrations. One lidocaine concentration was obtained within each of the three intervals following initiation of lidocaine infusions: I1 (1–6 h), I2 (6–12 h), and I3 (>12 h). Patients were categorized into 4 groups: (a) short-term infusions (<24 h) without CHF, (b) short-term infusions with CHF, (c) long-term infusions (>24 h) without CHF, and (d) long-term infusions with CHF. The mean prediction errors (range −0.60–0.27) included zero (95% confidence limits) in all groups and suggested no bias. Forecasts of the I3 lidocaine concentrations were consistently more precise [lower mean absolute errors (MAE) and root mean squared errors] using the lidocaine concentration obtained during the 6–12-h interval (I2) than when the lidocaine concentration obtained at the earlier interval (I1) was used. The MAE was reduced by 20–40% when a single lidocaine concentration obtained during I2 was used as compared to I1. Precision was only slightly improved with the use of two lidocaine concentrations. We conclude that this Bayesian algorithm is unbiased and delivers acceptable precision in forecasting lidocaine concentrations.