Abstract: Grasslands are the largest ecological barriers for the northwest agro-pastoral zone of China. Grassland classification is significant for ecological environmental conservation. This paper defines a grassland classification technique based on the remote sensing technology in Gansu Province, as the typical region for northwest agro-pastoral zone. High temporal resolution of Moderate Resolution Imaging Spectroradiometer (MODIS) is used to construct temporal profile of enhanced vegetation index (EVI) during the grass growth period (April to September). Typical temporal profile of EVI for each grassland type was constructed based on synchronous field samples. The study area is divided into a few of natural subregion based on elevation and the Yellow River division. Decision tree was established with similarities between each pixel and typical temporal profile, and classification was completed in each natural subregion. Based on the 1︰500 000 scale maps of China’s grassland resources, the validation process indicated that overall accuracy of classification results was 71.41%, and kappa coefficient was 0.66. The area of each grassland type was also close to that in the sample atlas, which proved that MODIS EVI was effective for grassland classification. As MODIS images are free and suitable for large area monitoring, it is possible to conducted low-cost, high precision and macro grassland classification.