Abstract: Remote sensing image segmentation is an important procedure of object-oriented information extraction of high-resolution.The mean shift segmentation algorithm has been applied to many segmentation procedures since it has good effects and perfect efficiencies.Firstly,we give the principle of mean shift segmentation.Then,we present the method to improve the segmentation algorithm and implement a multi-scale segmentation.Finally,the algorithm was evaluated on the basis of experimental comparison and analysis.The results confirm feasibilities of the mean shift segmentation algorithm.