A fuzzy Markov random field (FMRF) model is established and a new algorithm based on FMRF for image segmentation proposed in this paper. This algorithm simultaneously deals with the fuzziness and randomness for effective acquisition of the prior knowledge of the images. A conventional Markov random field (CMRF) serves as a bridge between the FMRF, obviously a generalization of the CMRF, and the original images. The FMRF degenerates into the CMRF when no fuzziness is considered. The segmentation results are obtained by fuzzifying the image, updating the membership of prior FMRF based on the maximum posteriori criteria, and defuzzifying the image according to the maximum membership principle. The proposed algorithm can effectively filter the noise and eliminate partial volume effect when processing the degraded image to ensure more accurate image segmentation.