Maximum scatter difference method is a good improved algorithm for the classical Otsu method. In order to improve the algorithm’s segmentation accuracy when the image is interfered by the noise or uneven illumination,a new maximum scatter difference image thresholding segmentation algorithm base on two-dimensional histogram oblique (TOMSD)is proposed. This algorithm considers the impact of between-class divergence and within-class divergence on the image information classification,and takes advantage of image space information to improve the ability of anti-noise. The fast recursion algorithm is given to speed up computational time. The results show that TOMSD has more accurate segmentation effect and a better anti-noise property than the two-dimensional histogram oblique Otsu method and two-dimensional histogram oblique maximum entropy method,and the running time is less.