Triangulated Irregular Network (TIN) has been widely used in the simplification of terrains represented as height field because of its predominant capability. During the generation of TIN, selection of input points is the key to the approximation quality of simplified terrains. Usually, the input point is selected according to the importance measure on each point, so research on the calculation of importance measure has attained more and more attention. Based on analysis and comparison of traditional importance measure, a new criterion of importance measure based on local entropy is put forward. The results demonstrate that the local entropy criterion gives a performance that is better than any traditional method. In addition, it can effectively conquer the "short sight" problem in previous methods.