Abstract: In the ultrasonic B-scan inspection for railway wheels, defection region is to be isolated from the B-scan image on the basis of image segmentation. According to the deferent features of B-scan image at vertical and horizontal directions, a new method is introduced in this paper to solve the B-scan image segmentation as a result of improving the traditional region growing method. In the new method, the pixels with maximal color value are picked out as the seed pixels, and then different growing principles are used in the region growing method at vertical and horizontal directions, in which the principle of wave valley threshold is used at the vertical direction, and the principle of correlation coefficient threshold is used at the horizontal direction. By the experiments of ultrasonic B-san for railway wheels, it shows that the new region growing method behaves well and acquires a satisfied result of defection region segmentation.