Surface hardness is an important index for evaluating the quality of surface processing. Ultrasonic rolling and extrusion technology plays a very significant role in surface strengthening. Taking 42 CrMo bearing steel as the object,the range analysis was carried out on the orthogonal experiment results of ultrasonic rolling extrusion,and the significance of process parameters on surface hardness was obtained. The reliability and accuracy of BP neural network model and stepwise regression model were compared and analyzed in sections by using k-fold cross validation method. This process fully considers the fitting ability and prediction ability of the model. The results show that the validation error range and average error of the stepwise regression model are smaller,and the prediction accuracy of the model is higher. Finally,the established prediction model of surface hardness has strong overall and coefficient significance,which can be applied to the optimization and improvement of surface quality in ultrasonic rolling extrusion.