///Spectral feature is the physical basis of rock identification. In order to remove the rock spectral noise, the spectra including common 15 rock samples belonging to 10 rock types are pretreated by averaging, resampling, smoothing, and fitting the value of water vapor absorption band. The continuum removing methods are used to obtain absorption-band parameters of spectra. The rock spectra are classified into four types by two-dimensional image analysis. From first type to last type, the spectra prove the gradually shallow absorption depth, the decreasing area, the gradually narrowing width and the increasing number of absorption peaks. Iron and pelitization alteration phenomenon are obvious. The classification results verified by cluster analysis are of better correspondence.