The surface quality of grinding has a direct influence on the performance of face gears. Based on the orthogonal experiment results of face gear grinding, the method of analysis of variance has been adopted to obtain the influence law of grinding wheel speed ns, work-piece feed speed vw and grinding depth af on grinding surface roughness Ra, thus proposing a parameter optimization scheme. By using multiple linear regression method, a prediction model of the relationship between surface roughness and grinding parameters has been established, with the experimental results testifying its better accuracy. An analysis has been made of the metallographic structure of the grinding surface of the gears, thus obtaining results of the experimental matrix and the residual austenite volume of the tooth surfaces. A prediction model of residual austenite on grinding surface can be established by adopting quadratic response surface method. With the residual austenite volume to be predicted, the model exhibits an obvious significance in testing.