The complexity and diversity of the texture of the surface defects on the surface of strip steel, as well as the false defects in the background texture, have brought great difficulties to the feature extraction and recognition of the strip steel surface defects. Therefore, this paper presents a new method for the selection and identification of steel strip surface defects. First of all, through the inhibition of anisotropic diffusion algorithm for suppression of false defect on strip surface; secondly, selection of surface defect features using PRelief method in this paper, compared with the traditional Relief method, this method considers the different dimensions of feature correlation. Finally, using the feature set and SVM kernel classifier to classify and recognize the surface defects of steel strip. The experimental results show that the proposed method can extract the features of strip surface defect pairwise independence and robustness, and for scratches, bumps and folds, stains and other different types of defects, this method compared with the traditional method can get higher recognition rate.