Abstract: Based on the dimension disaster caused by PCA in the image processing of metal fracture, a recognition method of metal fracture images based on 2DPCA is proposed. In the proposed method, 2DPCA is based on maximizing inter-class divergence, and its covariance matrix is directly constructed from the original image matrix. At the same time, the proposed method is compared with the PCA recognition method. The experimental results show that the proposed recognition method is less computable and the higher recognition rate than the PCA recognition method. In addition, it is very important to choose the suitable feature space dimension. When the feature space dimension is too small, the image information is imperfect and the recognition rate is low. When the feature space dimension is too large, the image information is redundant and the calculation amount is increased.