This study used the normalized difference vegetation index (NDVI), in conjunction with other ancillary indices, a digital elevation model (DEM), and the multistep unsupervised classification method to classify grassland in Gansu Province and the Qilian Mountains in China. The results showed that the overall accuracy of vegetation type reached 88.79% and that of grassland coverage level reached 87.23%. The ancillary indices suitability analysis revealed that meadow was distributed mainly in zones where the normalized difference moisture index (NDMI) varied between −0.64 and −0.4, whereas for steppe, it varied between −0.55 and −0.32. Grassland with a different coverage level was mainly distributed in zones where the normalized difference soil index (NDSI) varied between −0.20 and 0.25. To demonstrate the usability of these two indices, the maximum values of NDVI, NDMI, and NDSI and the DEM were used in the decision tree classification method for grassland. The results achieved relatively high kappa coefficients of 77.09% for vegetation type and 65.29% for grassland coverage level. Based on these results, it can be concluded that it is rational to apply the multistep unsupervised classification method and the selected indices for regional-scale grassland identification when a priori information is scarce, expensive, or unsuitable.