A new method of synthetic aperture radar (SAR) image segmentation is p roposed,which simulates the mechanism of visual perception with multiscale auto-regressive an d moving average(MARMA) model,and Markov random field (MRF) models.Firstly,the chara cteristics and mechanisms of the visual perception system are analyzed.Secondly,th e spatial scale and orientation perception mechanisms are simulated by constructing pyramid stru cture and MARMA model for SAR images.Then,the multiscale segmentation is achieved by the visual mechanism,the MRF model on different sca les,and the improved simulated annealing algorithm.Experimental results on SAR images demons trate that the proposed method has good performance for SAR image segmentation.