Abstract: Based on the proposed gray-gradient co-occurrence matrix (GGCM) and the clustering analysis for the weld defect, magnetic flux leakage (MFL) images of the cylinder defects in the weld, the rectangular slot defects in the weld, and the rectangular slot defects in the heat affected zone were respectively taken as the research objects to further identify the weld defect. The features for these three state MFL images extracted by GGCM were passed to hierarchical clustering, and the characteristics selected by this clustering were analyzed by using k-means clustering method. Results show that the recognition rate of weld defects by these three kinds of methods is more than 93.33%. And test results verified the feasibility of this method in different types of weld defects and the weld defects in different locations. It also showed that identifying weld defects of the same type in different locations was much easier than that of different type defects in the same position.