Objective To investigate the accuracy of Bayesian fitting for predicting aminoglycoside concentrations using Abbottbase pharmacokinetic systems program (PKS). Methods Data come from 53 patients treated with gentamicin and tobramycin iv injection in Goettingen University Hospital. Bayesian fitting of PKS was used to predict the peak and trough levels afterwards for 3 days. The prediction bias and precision were evaluated by mean prediction error (ME), mean squared error (MSB), root mean squared error (RMSE) and their 95% confidence intervals: All the predicted blood levels were compared to the measured levels using student paired Mest during the 24, 48 and 72 h of treatment. Results Three hundreds three aminoglycosides concentrations were available to predict, ME was -0.05 and - 0.28 mg·mL-1 for peak of 24 and 48 hours post-infusion, respectively. Their 95% confidence intervals included 0, showing no bias. All MSE, RMSE for peak of 24 and 48 h were small and within the clinical accepted range. No significant difference were found for peak levels between predicted and measured levels on 24 and 48 h (P>0.05). However, there are significant difference in 72h and all trough levels (P<0.05). Conclusion predicting peak concentration of aminoglycosides using B ayesian fitting of PKS will be acceptable with two feedback concentrations. It seems able to help clinician adjust individually aminoglycosides dosage easily. However, it may be limited for prediction of trough concentration.