Abstract: Forging temperature is one of the most important process parameter in the forging process of new mechanical bearing steel. In order to optimize the forging temperature of new mechanical bearing steel, a neural network optimization model with 3×18×1 three-layer topological structure was constructed. Selecting the bearing steel brand number, initial forging temperature and final forging temperature as the input layer parameters, wear resistance as the output layer parameter, tansig function as the implied layer transfer function, and purelin function as the output layer transfer function, learning training and prediction validation were performed for the neural network optimization model. The results show that the relative training error of the model is between 3.03% and 6.67%, and the average relative training error is 5.00%; the relative prediction error is between 3.13% and 5.41%, and the average relative prediction error is 4.22%. The model can accurately reflect the influence of the brand number, initial forging temperature and final forging temperature on the wear resistance of the steel, and the model has strong prediction ability and high prediction accuracy.