Abstract: A BP neural network model is established between preparation process and magnetic properties by using uniform design solution and principal component analysis.This model is used to optimize the preparing process of bonded NdFeB permanent magnet.The effects of single element or the interaction among elements on magnetic properties are discussed respectively.The precision of the forecasting model is higher and the largest relative errors between the predicted and measured results are 1.85% for Br, 1.28% for H_cj, and 1.47% for (BH) _m respectively.