Surface roughness of workpiece is a crucial parameter among integrity indexes, and it is also one of the important factors to measure the grinding quality. Predicting the grinding surface roughness accurately has great significance for selection of the grinding process parameters in a more rapid and reasonable way.The grinding processing data were acquired through actual grinding experiment. In order to adapt to the RBF neural network learning, the data were normalized. At the same time, cyclic algorithm was used for choosing optimal number of neurons in the hidden layers.Eventually, the grinding surface roughness prediction model was established based on RBF neural network. The MATLAB simulation results show that the prediction model has high accuracy, and can provide reliable data for surface roughness prediction research.