The predictability of serum vancomycin (VCM) concentrations by means of the Bayesian method was evaluated to establish whether the method can be used to select safe and effective VCM treatment regimens for individual patients. Serum VCM concentrations at the trough and 2 h after the end of infusion (peak) were measured. Pharmacokinetic parameters were calculated for VCM dosage regimens based on a two-compartment model with the Bayesian method, using the Japanese population pharmacokinetic parameters estimated by Yasuhara et al. (1998). The predictive performance for serum VCM concentrations and the dosage regimens were analyzed using two points of serum VCM concentration in 41 patients whose serum creatinine and age were in the ranges of 0.4—4.6 mg/dl and 24—92 years, respectively. Although the predicted values for trough and peak VCM concentrations were slightly lower than measured VCM concentrations, the predictive performance was generally good. There were no differences among the groups classified by serum creatinine or age. An examination of predicted data that differed markedly from the measured serum VCM concentrations indicated that a larger difference in volume of distribution at the steady state (Vdss) calculated from serum VCM concentrations at the beginning and revision of dosage regimens resulted in a poorer correlation of predicted values and measured values. This finding indicates that therapeutic drug monitoring should be conducted frequently, and the dosage regimen revised accordingly, in the case of patients who may have a change of Vdss of VCM, for example, due to a complication such as heart failure or edema.