Set partitioning in hierarchical tree (SPIHT) coding is a widely applied and very effective technique for image compression at present. Based on this algorithm, in this paper we offer the hierarchically weighted method after wavelet transform, which processes the wavelet coefficients by multiplying different weighted coefficients according to the different bit rates, so as to alter the relation of subbands of each level, specially between the coarse signal and detail signal. The practical tests show that this improved algorithm possesses a better compression performance and the peak signal-to-noise ratio (PSNR) is higher about 0.2-0.7dB than the primary algorithm.