Segmentation is a fundamental issue in synthetic aperture radar (SAR) image analysis. When contrast results against the true landscape, multi-polarization SAR images can provide more information, and obtain more correct segmentation than only single polarization SAR image. This paper presented a new SAR image segmentation method based on the multi-polarization SAR images fusion. The method built the whole image segmentation gradually by the segmentation algorithm framework based on ratio of averages (ROA), and the fusion processing was applied at the ROA gradient computing and the region merging step. A new method for ROA computing was proposed to generate gradient for each polarization image, and the first fusion processing was applied to build the gradient image by all of the gradients of multi-polarization SAR images. Then the initial segmentation of the gradient image was obtained by watershed algorithm. Finally, the final segmentation was provided by region merging using multi-polarization information as the second fusion step. Experimental results on the NASA/JPL multi-polarization SAR image showed that the method can improve the segmentation and have good performance.