Abstract: A novel image coding algorithm based on coefficients of interest is proposed in terms of the special needs of remote-sensing images. When coding rate is given, the entropy estimate is calculated from the coefficients of the given image, so that the last coding bit-plane can be derived when the image is coded normally. The difference of the algorithm from the one for region-of-interest algorithm lies in that the region and the scaled value need not be preappointed, and the most important is that all the coefficients in bands are not scaled up, while the coefficients of interest and the scaled value are deduced from the given image in the algorithm. This algorithm can be thought as the generalization from the region-of-interest algorithm, which is smarter and more valid than the region of interest coding algorithm. The simulation and the experimental results show that the small targets in the remote-sensing images can be saved validly while the whole image quality decreases a little.