Abstract: This paper presents a defect detection algorithm to detect the defects of welding image successfully. This algorithm is based on the gradient mask which is derived from piecewise fitting method using quadratic function and contour tracing method according to the human vision characteristics. By incorporating statistic recognition with fuzzy recognition and analysing the shapes and features of the welding defects, various welding defects are recognized and classified. Individual and comprehensive quality classifications are made for the welding defects according to the National Standard. The experimental results show that the proposed method performs satisfactorily for the defect detection and quality classification with higher inspection speed.