To improve the robust of the image hash method, a new style image hash algorithm based on BP neural network was proposed. Firstly, the image pixel matrix and a constructed function was used to train the back propagation BP neural network, and then the low-frequency components obtained by the discrete wavelet transform was used to composes the matrix, the image hash sequence was generated by the well-trained back propagation BP neural network in the end. The experimental results indicate that the proposed hash method is robust against content-preserving modifications (such as JPEG compression, image filtering) and has good application value.