Abstract: The evaluation databases for the WSR-88D hail detection algorithm have been built by using hail observation data of 504 hail prevention spots in Guizhou and Doppler radar data of Guiyang during 8 of severe hail cases from 2005 to 2006, filtered by specific conditions including definition of severe hail, observation range along the cell track, and time-window methodology, etc. These conditions make the databases provide a more accurate picture of algorithm performance. The algorithms are evaluated using the probability of detection (D_PO), false alarm ratio (R_FA), and critical success index (I_CS) statistics.It shows that POH (probability of hail) threshold of 50% get the highest R_FA, and different POH thresholds get similar I_CS, suggesting that it is unreliable to use POH as the only parameter for hail detection in Guizhou region. The difference of climatology between Guizhou area and central Switzerland where the initial POH curve is derived is the crucial cause why POH algorithm becomes unreliable and gets higher R_FA.Assigning POSH (probability of severe hail) 30% leads to the highest I_CS score in Guizhou region, but this threshold does not always get the best performance in these 8 severe hail cases. The difference of WTSM (Warning Threshold Selection Model) in different climatic region is the main cause why the default POSH algorithm gets a bad performance. An improved WTSM will predict the optimum I_SH threshold for each day more accurately. It will help ensure that the POSH threshold of 50% always corresponds to the largest possible I_CS every day. The re-evaluation of the improved POSH algorithm shows that it has decreased the hail detection R_FA, and gets a higher performance of severe hail detection in Guizhou.