Hyperspectral technology has received great attention for estimating aboveground plant biomass. Corrections using ground spectrometer data have been proposed for more accurate integrated assessments of aboveground plant biomass. In this study, we sampled the aboveground plant biomass and the corresponding hyperspectral data from three types of alpine grassland ecosystems (alpine desert, alpine steppe, and alpine meadow) from the Qinghai-Tibetan Plateau to improve the biomass estimation accuracy. We used partial least squares regression and the vegetation indices (VIs) (using all two-band combinations (601*601combinations in all) involving 601 narrow bands between 400nm and 1000nm), and the optimal band combinations were used to calculate four most commonly used VIs. We found that the VIs are better than the least square method for alpine meadow, and that the most effective combinations were shown in the combinations of reflectance at the red edge (rather than red) and near-infrared bands(728nm,764nm). For alpine steppe and alpine desert, however, partial least squares regression is better than VIs. While the most important band for alpine steppe is between 600nm and 700nm in the red spectrum, the most important band for alpine desert is around 400nm. Our results can provide a theoretical basis for accurately estimating the aboveground plant biomass of different grassland types in the Qinghai Tibet Plateau.