Landscape indices are effective tools to explain the spatial structure and patterns of ecological landscape including area/density, shape, core area, isolation/proximity, contagion/interspersion, and connectivity. More than 100 indices have been developed and an increasing amount of research explains changes in urban spaces using the indices. However, landscape indices have a high level of sensitivity to the scale of analysis - grain size and extent. If the scale sensitivity of indices is not considered, the research may produce inaccurate results. This study examines the scale sensitivity of landscape indices to find relatively stable indices in the complex geographical features of Korea. The scale sensitivity was analyzed using 20 categories of grain size and 41 categories of extent change. Landsat TM and ETM+ images of five years - 1985, 1991, 1996, 2000 and 2003 - were used, and 54 class level indices mounted on the FRAGSTATS program were examined. The results are as follows: First, according to the analysis of the scale sensitivity, 19 out of 54 class level indices were found to be stable to scale change. Second, the scale sensitivity was closely related to the green area ratio, and the typical threshold of change was $40{sim}50%$. Third, among the 16 indices which were frequently used in the research in Korea, only 6 indices were relatively stable to the scale change. These results can be an effective basis for the selection of indices in the landscape ecology research in Korea.