The accurate classification of polarim etric synthetic aperture radar (PolSAR) images is a challenging task because o f the existence of the speckle noise resulting in many false alarms.By combining the capability of the distribution in fitting different clutter regions in the PolSAR image and the capability of the Potts Markov random fields of modeling the contextual class information between neighb oring pixels,a new unsupervised classification algorithm for PolSAR data is proposed based on maxim um a posteriori (MAP) criterion.Firstly,the conditional iterative mode algorithm and the Metropolis sa mpling algorithm are utilized to refresh the class type of each pixel by iteratively resolving the objective func tion which is established by the MAP classification criterion.Secondly,at each iteration step,to get more accurate cl assification result,the distribution parameters are estimated by using the method of matrix log-cumulan ts which is based on the Mellin transform.Finally,the final class type of each pixel is the one which appears mo st times in the iteration steps.The experiment utilizing an NASA/JPL/AIRSAR polarimetric SAR image demonstrates that the proposed algorithm gets more accurate classification result than the Lee method,and the distribution fi ts the clutter of the PolSAR better than the Wishart distribution,K distribution and G^{0distribution}.