Abstract: The accuracy of land cover classification is affected by the low utilization efficiency of the TM image spectral characteristics.Aimed at this issue，a classification method is proposed based on the optimal band combination.Taking the TM image in 2006 in Zhanggong district of Ganzhou as a study goal，the optimum index factor（OIF）is firstly calculated by the spectral characteristics of remote sensing image and the relation of bands.Then，the revised normalized difference vegetation index（RNDVI）is introduced according to the topographic feature in study region，and the original image is analyzed by principal component analysis(PCA).Through comprehensive analysis on the optimum index factor，the revised normalized difference vegetation index and three former principal components，it is concluded that the optimal band combinations are PC3，RNDVI and Band1.Finally，with combination of supervised classification and unsupervised classification，the composite remote sensing image is classified，obtaining that the overall classification accuracy is 86.2375% and the kappa factor is 0.8253.

Key words: The optimal band combination, Optimum index factor（OIF）, Revise normalized difference vegetation index（RNDVI）, Principal component analysis（PCA）