Remote Sensing Image Classifiction is always a difficult problem.It’s very important to use proper algorithms for different images.Not only precision but also efficiency must be considered.By researching in the relations between the initial means of clusters and the efficiency of clustering

we proposed a method of computing the initial means of clusters for K-Means Clustering.It is proved to be more efficient than the（K-Means） Clustering module of ENVI when it’s used in Remote Sensing Classification.