Abstract: The algorithm of image noise reduction and enhancement is designed according to the characteristics of weld radiographic image, in the case of various unfavorable factors such as bad contrast ratio of weld defects, illumination asymmetry and many textures; the algorithm for extracting the weld (defect) is designed, based on edge inspection under weld background condition. Through analyzing weld (defect) features, defect feature parameters are selected, the fuzzy neural network model used to recognize weld (defects) is developed, and the methodology of constructing membership are introduced. From experiments, it was successful for image preprocessing and defect extraction. The recognition algorithm could raise the recognition ratio based on fuzzy boundary pattern classification, which is better than classification recognition method for weld defect recognition.