The traditional lifting wavelet transform cannot be effectively reconstructed non-horizontal and non-vertical high-frequency information of remote sensing images, which results in these high frequency wavelet coefficients to be still salient relatively and reduces the coding efficiency of remote sensing images. A new lifting wavelet scheme based on direction optimal model called direction optimal lifting wavelet (DOLW) is proposed. The new algorithm first designs a directional prediction model based on gradient to obtain optimal transform direction of lifting wavelet. It executes the directional lifting transform in the direction of the first vertical along the optimal transform direction. The edge and texture energy can be weaken on the non-horizontal and non-vertical direction of the high-frequency subband of the remote sensing image. The new algorithm uses sampling function interpolation to predict the value of sub-pixels. The experimental results for these remote sensing images show that, compared with traditional lifting wavelet transform, the new algorithm improves the peak signal-to-noise ratio (PSNR) and the subjective quality of the reconstructed images significantly. So the new algorithm has important value for the remote sensing image compression and coding in the future.