For leaf spring of clutch cover assembly, the punch wear in the blanking process was simulated by Deform-3D, and the punch wear was predicted based on orthogonal experiment and BP neural network. Then, blanking clearances, fillet radius and blanking speed were taken as the input layer, the wear depth of punch was taken as output layer, and BP neural network with three layers of 3-12-1 was established, and the maximum error of the prediction was 1.14% by training of BP neural network. Then, the performance of BP neural network was tested based on simulation data of orthogonal experiment, the error between BP neural network and simulation value reached 2.09%. Furthermore, the predicted value of BP neural network was verified by the progressive die of leaf spring, and the error between BP neural network and experimental value was 8.25%. Therefore, the accuracy of BP neural network predicting the punch wear of leaf spring was verified.