The distributing of China’s grassland is abroad and the status of grassland degradation is in serious condition. So achieving real-time and exactly grassland ecological monitoring is significant for the carbon cycle, as well as for climate and on regional economies. With the field measured spectra data as data source, hyperspectral remote sensing monitoring of grassland degradation was researched in the present article. The warm meadow grassland in Hulunbeier was chosen as a study object. Reflectance spectra of leaves and pure canopies of some dominant grassland species such as Leymus chinensis, Stipa krylovii and Artemisia frigid, as well as reflectance spectra of mixed grass community were measured. Using effective spectral feature parametrization methods, the spectral feature of leaves and pure canopies were extracted, so the constructive species and degenerate indicator species can be exactly distinguished. Verification results showed that the accuracy of spectral identification was higher than 95%. Taking it as the foundation, the spectra of mixed grass community were unmixed using linear mixing models, and the proportion of all the components was calculated, and the errors were less than 5%. The research results of this article provided the evidence of hyperspectral remote sensing monitoring of grassland degradation.