Objective To perform a head to head comparison between three generations of Partin tables, namely from 1997, 2001 and the last updated version of 2007. Material and methods The external validations were based on clinical and pathological data of 687 consecutive patients undergoing radical prostatectomy for clinically localised prostate cancer between 2003 and 2008. Three versions of the Partin tables were compared for their accuracy and performance to predict final pathological stage using receiver operating characteristic (ROC) curve and Loess plots analyses. Results Of the whole cohort, 76.2% of men were presented with organ-confined disease (OC), 17.0% had extraprostatic extension (ECE), 6.0% showed seminal vesicle involvement (SVI) and 1.2% had lymph node involvement (LNI). The area under the receiver operating characteristic curve (AUC) of the Partin Tables 1997, 2001 and 2007 was 0.731, 0.727 and 0.722 for OC; 0.671, 0.662 and 0.650 for ECE; 0.795, 0.788 and 0.779 for SVI as well as 0.826, 0.786 and 0.746 for LNI, respectively. Conclusion All three generations of the Partin tables showed a good accuracy to predict OC, SVI and LNI. However, the predictive accuracy for ECE was only modest. Overall, the newer versions of the Partin tables could not exceed the version of 1997 in their predictive accuracy for any pathological stage and they failed to demonstrate a clear advantage. Our results underline the necessity to perform external validations before the implementation of a new predicting tool.