Among the remote sensing images segmentation methods, Gaussian mixture model (GMM) is the widely used image model. Gaussian-Rayleigh mixture model (GRMM) is proposed, and it may be more suitable for remote sensing image modeling. The difference between classical GMM and GRMM is introduced. The modeling results of GMM and GRMM to the images are compared. The comparison data shows that the GRMM has less distribution errors than the GMM when modeling the images. The entropy-max method is utilized to determine the optimal class number. The Markov random field (MRF) and a new potential function is employed to segment the images. The iterative conditional model (ICM) is used to calculate the maximum posteriori probability. Three remote sensing images are utilized in the experiment, the fitting and segmentation results of GMM and GRMM are compared in the experiment process. The data and segmentation results show that the proposed method is more effective.