Abstract: The enormous volumes and valuable applications of MODIS multi-spectral images make it necessary to be compressed losslessly with effective methods.To solve this problem,an optimal linear prediction and band ordering are adopted to exploit abundant spectral redundancy,and SPIHT algorithm is used to eliminate the spatial redundancy of MODIS data.The optimal inter-band prediction sequence is specified by band ordering.Except the first band,only residual error images of other bands need to be encoded after band prediction.To avoid information loss,the optimal linear predictor is improved by rounding the prediction coefficients into integers.The experimental results show that our method has more compression capabilities.