The performance of the Bayesian feedback method for predicting lignocaine dosage developed by Vozeh et al. was evaluated in 53 Japanese patients, treated for either suppression or prophylaxis of ventricular arrhythmias, with respect to accuracy, precision, efficiency, and safety. The lignocaine serum concentrations at 12 h (C12) and 24 h (C24) after continuous infusion was started were predicted and compared retrospectively using the early concentration at 2-4 h (C2-4). The mean of the prediction error (PE) and the root mean-squared error (RMSE) were used as a measure of accuracy and precision. The 95% confidence interval of PE for C12 included zero, but those of PE for C24, whether C2-4 was used alone or in combination with C12, were less than zero, meaning that the prediction of C24 was significantly biased. The RMSE for C12 (13.2%) was less than that for C24 (16.2%), but the difference was not statistically significant. The fraction of measurements lying within the statistically approximate 68% prediction interval (+/- 1 SD) was used to evaluate the estimate of the SD of prediction. The fraction of measurements inside the prediction interval was significantly larger than the expected 68% for both C12 and C24 (p less than 0.05), showing that the estimate of the size of the PE is biased on the safer side. The clinical effects of lignocaine were also evaluated from the viewpoints of suppression of number of ventricular premature beats, prevention of more severe grades of arrhythmias, and toxicity. At least the appearance of toxicity, which four patients exhibited, was related to the lignocaine serum concentration (greater than 6 micrograms/ml).