The SPOT multispectral data, taken on Dec. 31, 1986 and Jan. 15, 1988 were utilized in this study. The data have been processed; such as atmospheric correction, brightness correction, geometric corrcction, etc, Then use six different change detection methods to analyse the variations of burnt-over area, vegetation community and land-slide changes. The purpose of this study is to find an optimal detection method for Taiwan's mountainous forest area. The results indicates that spectral vector change detection is the optimal method and can be applied on the change detection of forest land-cover in Taiwan.