We assessed the performance of a predictive algorithm for dosing aminoglycoside antibiotics in 75 pediatric patients and Bayesian feedback in 36. The absolute errors for peak and trough concentrations were 1.83 and 0.80 mUg/ml, respectively, which seem clinically acceptable for most patients. However, the algorithm had significant negative bias for both peaks and troughs. Implementation of Bayesian feedback eliminated bias in a second set of concentrations and significantly decreased its magnitude for both peaks (p= 0.028) and troughs (p= 0.005). This method may allow more accurate dosing of aminoglycoside antibiotics in pediatric patients, though it would most likely be improved by better definition of population parameters and their variability.