To research on the roles of multiple features in the SAR image segmentation, this paper presents a SAR image segmentation method with weighted curvelet features in the Bayesian framework. Firstly, curvelet transform is used for the SAR image to extract multiscale spectral features of every pixel, then pixels' feature vectors can be formed. To research on the roles of extracted multiscale spectral features in the SAR image segmentation, a weight is assigned to a component in the pixel's feature vector. Then, an image with feature weighted can be defined by the feature and weighs, and its image domain is partitioned. On the partitioned image domain, a SAR image segmentation model based on weighted curvelet features is built in the Bayesian framework. Further, Markov Chain Monte Carlo (MCMC) and expectation maximization(EM) algorithms are used to segment image and estimate the weight values. Finally, the proposed method and four comparison methods are used to segment the SAR image. The quantitative and qualitative results illustrate that the proposed method can not only automatically determine the roles of multiple features in the segmentation procedure, but also improve the segmentation accuracy effectively, and the proposed method has strength in the multi-feature segmentation of the SAR image.