This paper proposes a new SAR image segmentation method based on graph and gray level reduction in Independent component analysis(ICA) space. Firstly,according to the grayscale information of SAR image, effective use of gray level reduction for initial segmentation can group the pixels with same or similar values to the same homogeneous region, which can address the problem of over-segmentation. Secondly, the features of regions are extracted in ICA space, and then the similarity degree can be calculated by Euclidean distance. The initial regions are merged in fully connected graph based on minimum spanning tree in ICA space. The process of region merging is divided into two phases; the first phrase is merging the different regions with the largest similarity degree, the second will focus on updating the fully connected graph for iteration. Finally, experimental and comparative results on synthetic and real SAR images verify the efficiency of the proposed algorithm. This paper proposes a new SAR image segmentation method based on graph and gray level reduction in Independent Component Analysis (ICA) space. Firstly, according to the grayscale information of SAR image, effective use of gray level reduction for initial segmentation can group the pixels with Same or similar values ​​to the same homogeneous region, which can address the problem of over-segmentation. Where, the features of regions are extracted in ICA space, and then the similarity degree can be calculated by Euclidean distance. The initial regions are merged in fully connected graph based on minimum spanning tree in ICA space. The process of region merging is divided into two phases; the first phrase is merging the different regions with the largest similarity degree, the second will focus on updating the fully connected graph for iteration. Finally, experimental and comparative results on synthetic and real SAR images verify the efficiency of the proposed algorithm.