A new method for three-dimensional multispectral image data compression based on an improved Karhunen-Loève Transformation (K-LT)/wavelet transformation spectral feature coding vector quantification (WSFCVQ) is proposed. Three-dimensional Karhunen-Loève transformation/wavelet transformation/spectral feature coding vector quantification (SFCVQ) implement three-dimensional compression for three-dimensional multispectral image data with two-dimensional set partition embedded block (SPECK) and one-dimensional spectral feature vector quantification (SFVQ) coding, applying Karhunen-Loève transformation to exploit the spectral correlation in multispectral image data and using two-dimensional wavelet transformation (WT) to remove the spatial redundancy in the multispectral image data, and a high compression performance is got as well. Experimental results show: if Karhunen-Loève transformation/wavelet transformation/wavelet spectral feature coding vector quantification (WSFCVQ) method are compared with the method of Karhunen-Loeve transformation/wavelet transformation/improvement BiBlock zero tree coding (IBBZTC) and the method of Karhunen-Loeve transformation/wavelet transformation/fast speed vector quantification (FSVQ), the peak signal to noise ratio (PSNR) is enhanced by over 2 dB and 1 dB, the compression speed increased by 1.5 and 8 times respectively, and the total compression performance is greatly improved.