A new method is proposed for multispectral image data compression. The multispectral image data after Karhunen Leove Transformation (KLT) its spectral correlation and spatial redundancy are removed by using JPEG compression coding. A method is presented to decrease the bit flow to maximum degree. Experiment showed that this method can result in the compression ratio of 3.2~5.3 in the sixteen bands multispectral data to remove the spectral correlation without any loss. The total compression ratio of KLT+JPE can reach 103 (Q=50), PSNR being 34.0 dB. It is superior to the method of applying DCT to remove the spectral correlation.