A method that combines watershed segmentation with a modified competitive learning network (MCLN) is proposed for segmentation-based image compression. The watershed algorithm is used to segment a gradient image into several closed regions via region growing. The mean intensity for each region is then calculated for subsequent classification. After classification, vector quantization with MCLN is applied to these regions with different compression rates according to the importance of the regions to simultaneously preserve important features and reduce the size of images. The results indicate that the proposed method is promising in comparison with the generalized Lloyd algorithm and the MCLN algorithm.