The pharmacokinetics of vancomycin was investigated in adult ICU patients after the first administration and at steady state. Then the predictive performance of a two-compartment Bayesian forecasting program was assessed in these patients by using population-based parameters and three non steady state vancomycin concentrations as feedback information. Finally a prospective investigation was carried out to search potential covariates. At steady state, a significant decrease (around 30%) in clearance (CL) was observed, while creatinine clearance (CLcr) was stable and a significant increase (around 30%) in volume of distribution (Vss) was observed. A two-fold increase in elimination half-life was found. CL was weakly correlated with CLcr at onset of therapy and at steady state. The Bayesian program tended to over-predict vancomycin peak and trough concentrations. A larger mean prediction error and a poorer precision were observed when population-based parameter estimates were used (no feedback) compared to feedback prediction, but the differences were not significant. Mechanical ventilation and concurrent opioid therapy may be pertinent covariates of vancomycin pharmacokinetics. The current work has shown that vancomycin pharmacokinetics in ICU patients displayed a significant variability and a significant change in both clearance and distribution during the course of therapy. Further investigation is necessary to clarify these findings. Moreover, the use of the Bayesian forecasting PKS program in our patients led to a prediction with low bias but rather poor precision. This outcome highlights the need to implement a population modeling approach, to determine the vancomycin pharmacokinetic parameters and covariates in our ICU patients, and to apply this information to provide more accurate concentration predictions.