In response to the high data volumes of hyperspectral images and high data rates required for their transmission, many hyperspectral data compression methods have been researched in recent years . In this paper, a new compression algorithm for hyperspectral image is proposed, which pays much more attention to its peculiarity of much higher spectral resolution. The new method combines three-dimensional integer wavelet transform and vector quantization, as well as Said and Pearlman's SPIHT algorithm is extended to three dimensions. The principle ofthe new method can be explained in three steps. First, three-dimensional integer wavelet transform is performed on the hyperspectral image. Second, spectral vectors are formed on the basis of the orientation features of three-dimensional wavelet coefficients, thus the high spectral dependencies of hyperspectral image can be exploited simultaneously with wavelet tree structure. Third, an algorithm that extends SPIHT algorithm to three dimensions is used to encode the spectral vectors. In order to test the effectiveness of the proposed algorithm, 32 bands of an AVIRIS image are used for computer simulation. The results show that the SNR in the reconstructed images can reach more than 40 dB on average at 0.6 bit per pixel, which indicates the proposed algorithmis efficient for the compression of hyperspectral images.