In order to overcome the problem of irregularity of texture features when using the split-and-merge algorithm to segment texture images, the split-and-expand algorithm, a new quadtree-based texture segmentation approach, is presented in this paper. The initial and final regions to be segmented are determined by supervisedly classifying the texture samples of the different regions to be evaluated, and the precision of classification is treated as the weights of the texture features in the corresponding evaluated regions. Then the operation of spliting or expanding is performed according to whether the properties of consistency are satisfied or not. The method can effectively deal with the problem of inconsistent reliability of texture property in different evaluated regions, and the process of expansion can improve the precision of segmentation and avoid the instability of texture features in smaller regions. The results show that the split-and-expand algorithm can effectively segment texture images, and its precision of classification is superior to that of the split-and-merge algorithm.