Rangeland degradation at the Three-River Headwater Region(TRHR) has been a major concern to the public in recent years, especially toxic weeds. The application of hyperspectral remote sensing in detecting meadow composition and identifying degradation become a major tool for research. Using MESMA algorithm and Hyperion image combined with the measured hyperspectral data, information about edible grass, toxic weeds and bare land were extracted to classify the degree of rangeland degradation. The study area was divided into five levels of degradation, namely:no degradation, mild degradation, moderate degradation, severe degradation and extremely severe degradation. Compared with the measurement data samples on monitoring, the precision attained was 84.2%. This shows that hyperspectral remote sensing is effective in detecting rangeland degradation at TRHR.