Abstarct: The history of image segmentation can be traced back 40 years. It is one of the first steps in each image processing and machine vision system. Since the output of this block uses as the input of other stages in the system, the error of this stage is critical for the system error rate. Because of the growth of usage the real time processing systems, Computational cost is another important parameter in these algorithms. So it is very important to design a fast segmentation algorithm with low level error rate. Image segmentation problem could be surveyed in two branches: supervised image segmentation and unsupervised image segmentation. Here we will study some unsupervised thresholding baxsed image segmentation techniques. There are some approaches for gray level image segmentation, where thresholding baxsed algorithms are more useful because of their simplicity, stability and fastness. We proposed a new scheme for this purpose, in which correlative histogram is defined and then it is modeled by mixture of Gaussian functions. Optimal thresholding levels are determined from the model and then applied to the image. Experimental Results show the superior efficiency of proposed method especially for noisy images. Same algorithm is suggested for color image segmentation. In this method at the first step, hue histogram of the image is calculated and the mixture model is determined for it. Then optimal thresholding levels are determined and applied to the image. For each segment in the segmented image, saturation histogram is calculated and modeled by a new mixture model. Then threshold levels are determined from these models and applied to each of the segments for more segmentation. Experimental results show this method works well specially for images which affected by illumination noise. The final attempt is about texture segmentation. Each texture image has its own statistic characteristics. In the proposed method we suggested to use of the local mean and variance of the image. At the next step, the histogram of each extracted feature if determined and modeled by a mixture model. Optimal thresholding levels are obtained from these models and applied to the image. Experimental results show that the proposed method achieves good performance in texture segmentation problem.