In this paper, the General Regression Neural Network proposed to build a predicted model of the wafer probe yield that the key variables form the Wafer Acceptance Test (WAT). Using the actual situation of Taiwan DRAM production processes as an example. From the analysis results, it can get the 82 significant variables of the WAT, and the R² of GRNN prediction model is 0.73. After discussing with the manufacturer, the model was introduced into the actual process, and the results can effectively increase the yield and reduce quality cost in the DRAM manufacturing.