Medical zirconia ceramic （Y-TZP） is a good dental restoration material. To obtain good dental restoration performance， high manufacturing accuracy， especially surface roughness， is required. However， it is a hard， brittle material， which is difficult to machine. To improve the surface quality and processing efficiency of medical zirconia ceramic grinding， correlation analysis is conducted on the frequency bands of acoustic emission signals during the grinding process of medical zirconia ceramic. Twelve sets of characteristic values strongly related to grinding surface roughness in the sensitive frequency band signals of 840—850 kHz are extracted， and a random forest neural network with high prediction accuracy is constructed. Finally， medical zirconia ceramic grinding surface roughness is obtained. The maximum relative error of acoustic emission prediction is less than 8.37%， and the research results have great reference value for intelligent online monitoring of surface roughness in medical zirconia ceramic grinding.