The image threshold selection of skew and heavytailed classconditional distributions were studied. Due to the deviation of the meanbased method in classification estimation, the medianbased method is more reasonable in threshold selection. Based on the square distance symmetrical cooccurrence matrix, the region median was defined, and then using median classified statistics method, a new threshold approach was proposed based on the square distance symmetrical cooccurrence matrix, and the multithreshold segmentation algorithms was advanced. Compared with Otsu′s and square distance, the proposed method not only has prominent segmentation performance for the images of skew and heavytailed classconditional distributions, but it takes the more spatial statistical information on account, compared with medianbased Otsu′s thresholding, the extracted object information is more complete, and the edge is clearer. For the small object probability distribution images, this method also has better threshold segmentation effect. To illustrate the correctness and effectiveness, based on the groundtruth images, the misclassification error results show that the proposed method can obtain the minimum value.