Welding is an excellent attachment or repair method. The advanced industries such as oil, automotive industries, and other important industries need to rely on reliable welding operations; collapse because of this welding may lead to an excessive cost in money and risk in human life. In the present research, an automatic system has been described to detect, recognize and classify welding defects in radiographic images. Such system uses a texture feature and neural network techniques. Image processing techniques were implemented to help in the image array of weld images and the detection of weld defects. Therefore, a proposed program was build in-house to automatically classify and recognize eleven types of welding defects met in practice. The proposed system is tested and verified on eleven welding defects as follows; center line crack, elongated slag lines, cap undercut, lack of interpass fusion, lack of side wall fusion, lack of root penetration, misalignment, root undercut, root crack, root pass aligned, and transverse crack. It was found that only two welding defects are failed in a total 3 from 308 images in the overall recognition. The lowest classification percent was found in case of lack of side wall fusion defect (92.9%). The overall average discrimination rate results from a combined technique of texture feature and neural network are about 99%.