Abstract: Based on the fact that Bandelet transform can adaptively track the geometric direction of the image, the transform is hence introduced to the feature extraction of metal fracture image and a new nonlinear identification method of metal fracture surface based on it is proposed. In the proposed method, the Bandelet entropy of metal fracture surface is extracted by the Bandelet transform and used as the feature vectors, and neural network as a nonlinear classifier. The proposed method is compared with the traditional recognition method of metal fracture based on wavelet transform. The experimental results show that the Bandelet transform overcomes the deficiency in the wavelet transform in dealing with the metal fracture edge, obtains more recognition effect.