In this paper, we presented a practical and effective image compression system based on Wavelet Support Vector Machine (WSVM) with Morlet wavelet kernel for compressing still images. The algorithm combines WSVM learning with discrete wavelet decomposition technique. Compression is achieved by approximating wavelet coefficients at each subband separately using WSVM regression. Results demonstrate in comparison with JPEG, SPIHT and SVM with Gaussian kernel algorithm, the proposed algorithm increases compression ratio for a given image quality; conversely it gives better image quality for a given compression ratio.