Abstract: Currently, the detection of welding defects is difficult and inaccurate in the X-ray images with low contrast, so the automatic computer-aided defect identification cannot be used in practice effectively. Aiming at this problem, a methodology for the weld defect detection was proposed, which was based on calculating the square value of neighborhood subtraction, is presented. The sequent process contains morphological reconstruction, edge detection and local enhancement, and image fusion technique. Experiment results showed that the method was valid for the weld defect detection of the X-ray images with low contrast.