Image segmentation is a fundamental and challenging problem in image processing and often a vital step for high level analysis. The aim of image segmentation is to divide an image into different categories based on features, such as intensity, color, histogram or context, where ea ch pixel in the image should belong to one class an d only one class. For the image segmentation by the histogram thresholds, several methods have been proposed. How ever, image segmentation can be two-phase (two categories ) or multiphase (more than two categories), the num ber of categories becomes an important problem in this kin d of segmentation. And the segmented results accord ing to the thresholds whether or not consistent to the image i s also a problem should be considered. In this pape r, we use the bi-level threshold way to divide the image’s histog ram step by step until the total variance between t he histogram and the fitting curve match the stop conditions. Af ter these thresholds have been calculated, we use t he structure and contour characterization of the image to deal with the rough result of the segmentation. Experimental results and a comparative study with the other efficient and know n multilevel threshold methods over synthetic and r eal images have shown that the proposed method is consistent t o the image contents especially in nature images