Image segmentation plays an important role in image retrieval system. In this paper, a method for segmenting images based on SOM neural network is proposed. At first, the pixels are clustered based on their color and spatial features, where the clustering process is accomplished with a SOM network. Then, the clustered blocks are merged to a specific number of regions. Experiments show that these regions could be regarded as segmentation results reserving some semantic means. This approach thus provides a feasible new solution for image segmentation which may be helpful in image retrieval.