In this paper, the crack width of partly prestressing concrete square-section beam under the bending moment is explored by BP neural network. Firstly, by theoretical analyzing and testing BP neural network, the main effect factors that influence the crack width are found and other second-class effect factors are neglected. The optimized BP neural network is created, and more influential factors can be considered in the mode at the same time and the method of predit the crack width by BP Neural Network is discussed. Training and simulating with different samples show that the results are in accordance with the actual data well. It can be seen that it is feasible and precise to predict the crack width of partly prestressing forcing concrete square- section beam under the bending moment by BP neural network, and it is much better than using the calculation formula of the china's code for design of concrete structures