Anisotropic diffusion has good effect on reducing noise and preserving edge, in fact it's an iterative process. An important question is to decide when the process should be stopped. The stopping time is chosen by minimal mean square error, signal-to-noise ratio and minimal correlation coefficient based on the information of the original (ideal) images and noises. The entropy and mutual information in information theory is adopted to develop a criterion on stopping diffusion process. The performance of the different criterions is proved by experiments, the specialty and the applying condition is also discussed.