Abstract: Abstract: Grassland classification is a fundamental need of grassland science. It is also a challenge to develop a comprehensive grassland classification system because of the multivariable and multi-functional features of grassland ecosystem. The Comprehensive and Sequential Classification System of Grassland (CSCS), one of well known grassland classification systems, involves a hierarchy of three classification levels (class-subclass-type, class is the basic level) and is advanced in quantification indicators. However, there are at least two aspects need to be improved at the basic classification level of CSCS: 1) the grasslands are grouped into classes according to the data involving annual precipitation and accumulative temperature, which are collected from meteorological stations. These data reflect the near-surface atmosphere hydrothermal conditions instead of the actual habitat of grasses; 2) The data of precipitation and temperature from ground observation can only present the conditions within a small area, but they are used through extrapolation to a larger region. In order to resolve the problems, the areal data of land surface temperature and soil moisture are introduced by quantitative remote sensing as main data sources for the basic classification level of CSCS to replace the parameters of precipitation and atmosphere temperature from ground observation. In this paper, the MODIS land surface temperature product (MYD11A1, daily with 1km resolution) and MODIS land surface reflection product (MYD09GA, daily with 0.5 km resolution) of Gansu province in 2008 were used to invert soil moisture based on Thermal Inertia Model with the help of a Soil Moisture Inversion Platform (SMIP) developped from ENVI/IDL. Then, the annual accumulative land surface temperature (＞0℃Σθ) and annual sum of soil moisture were carried out, and then they were fitted with annual accumulative temperature (＞0℃Σθ′) and precipitation data from meteorological stations respectively. Thermal zones were determined by temperature and humidity zones by K-value (moisture index), and the grassland class was obtained by coupling thermal zones and humidity zones. Finally, the grassland types were verified through the field investigation and the accuracy assessment was tested with confusion matrix. The results showed that: 1) The grassland in Gansu occupies five thermal zones ( frigid, cold temperate, cool temperate, warm temperate, warm subtropical) and six humidity zones ( extrarid, arid, semiarid, subhumid, humid , perhumid); 2) There are 26 possible types present in Gansu province, and three grassland classes that cover the largest area in Gansu are warm temperate-arid warm temperate zonal semidesert (ⅣB11), cool temperate-arid temperate zonal semidesert (ⅢB10) and cold temperate perhumid taiga forest (ⅡF37), with the total area of the three classes is 17.83×106 hm2, accounting for 44.43% of the total grassland area in Gansu; 3) The geographical distribution of grassland type indicates significantly vertical zonality pattern: with the altitude decreasing, frigid series grassland, cold temperate series grassland, cool temperate series grassland, warm temperate series grassland and warm series grassland distribute successively from southwest to northeast, which fit the terrain of Gansu province. 4) Accuracy assessment shows: the overall accuracy of grassland classification is 70.11% and the kappa coefficient is 0.67. The research can solve the problem of transforming from scattered site data to regional polygon data in CSCS and the uncertain borderline in punctate data extrapolation, and provide a new approach to the utilization of CSCS, which can carry forward the practical progress of CSCS.