||Weld extraction is an important step in the recognition and classification of defects in radiographic testing (RT). To eliminate the effects of considerable noise, defect edges and the discrete grey-level distribution of each column in the RT images for extracting welds, a weld edge extraction method based on Bézier curve fitting (BCF) and medial axis transformation (MAT) is proposed in this paper. First, BCF is applied to fit the grey values of each column in the original image and obtain a continuous grey-level curve, by which considerable noise can be removed and the grey value of defects can be increased. Secondly, to extract a continuous weld edge from the fitted image, the threshold method with a new threshold selection strategy and edge connection are proposed. Finally, to get a precise weld edge from one that is several pixels wide, MAT is used to extract the skeleton of the wide edge; a weld edge connected by a single pixel can then be obtained. The experimental results show that the weld edge extracted using the proposed method is continuous and smooth and can correctly reflect the boundary of the weld region. Furthermore, this method can greatly remove noise interference and effectively extract the weld edge of an image with defects, compared with other similar methods.