Abstract: A magneto-optical sensor is applied to acquire magneto-optical images of tight-butt joint weld whose width is less than 0.1mm. Edge details are often missed and the accuracy of weld detection is influenced when traditional morphology algorithm is applied to detect the edge of micro-gap weld magneto-optical image. Thus, selecting structure elements of three different scales at each different four directions with an approach of multi scales and multi structure elements based on morphology is used to acquire the micro-gap weld edge information. Also, the multi-scale edge detection result is compared with those results using traditional wavelet transform and Sobel algorithm. Three groups of experiments are carried out under different magnetic induction fields to detect the weld center by using the approaches based on multi-scale morphology and traditional morphology, respectively. According to the results, the multi scales and multi structure elements approach can obtain the micro-gap weld center more effectively, which provides a testing basis for tight-butt joint weld recognition and tracking.