For automobile pedal beam, the springback in flanging process was predicted by combining numerical simulation technology with GRNN neural network. At first, the process of pedal beam flanging was simulated by software Autoform, and their springback angles were compared with the actual ones under the same parameters to verify the accuracy and substitutability of simulation results. Then, the springback angle data of each detection point with different parameters were obtained by the orthogonal test, and the GRNN neural network was trained by MATLAB. In order to ensure the accuracy of prediction, the multiple sets of spread factors were trained. Furthermore, when the spread factor is 0.1, the network has the best approximation performance and prediction performance, which is regarded as the final network model to test, and the maximum error between prediction results and actual measurement results is 4.3% which satisfies requirements of production. The results show that the GRNN neural network is of high efficiency and high precision for the springback prediction in sheet metal flanging.