Abstract: To study the change of land use in Qilian Mountains, Landsat TM/ETM+ images during 1986 to 2015 in Qilian Mountains were collected, and processed for time-series ground surface reflectance data by using relative radiometric correction methods. Combining spectral extension expansion method and classification and regression tree(CART)classification method, land-covers were classified into eight classes including forest, grassland, cropland, water, ice snow, gobi, sand and the other. Time-series Landsat images were used to classify the land types. The result shows that relative radiometric correction methods reduce radiation difference of time-series image. The classification result in 2012 was validated by the ground data, with an overall precision of 88.72%, and a Kappa coefficient of 0.86. During this period, the area of cropland, forest and grassland in the study area was declining, and the fragmentation degree of grassland was increasing. The area of gobi increased, and the desertification caused by vegetation degradation was more serious. Finally, the situation for land use in the research areas was discussed and suggestions for land use were proposed.