Partially stabilized zirconia （PSZ） is widely used in the aerospace industry and other fields due to its superior properties. Surface roughness is a key index to evaluate the grinding level of PSZ ceramics. In order to reduce the prediction error of grinding surface roughness， an acoustic emission （AE） prediction of grinding surface roughness of PSZ ceramics based on correlation analysis and convolution-bidirectional long short term memory （CNN-BiLSTM）neural network is proposed. The correlation between the eigenvalues and the grinding surface roughness values in different frequency bands in the grinding AE signals are analyzed， and the optimal sensitive frequency band and the feature matrix of the grinding AE signals are selected as the input parameters of the CNN-BiLSTM neural network to reduce the prediction error of acoustic emission of grinding surface roughness. The results show that the average prediction errors based on correlation analysis and CNN-BiLSTM neural network PSZ grinding surface roughness AE prediction is under 3.92%.