Abstract: Remote sensing technology emerged in the early 1960s and was widely used in grassland yield estimation. Estimation of the relationship between vegetation index and biomass on different scales was always used to complete the conversion from points to surface. Fourteen vegetation indices which were commonly used in grassland yield estimation were selected to establish regression models of vegetation index-dry weight and vegetation index-fresh weight in typical grassland in Baiyinxil, Xilinhaote, Inner Mongolia. The DCA analysis of 14 vegetation indices showed that: 1) It was feasible to use fresh weight and dry weight in remote sensing estimation of grass biomass, and that although dry weight was better than fresh weight, for reasons of experimental conditions, fresh weight had a wider range of applications. 2) The first and second axis of DCA analysis represent the effects of soil and atmosphere with soil the most important factor that affected vegetation index; DCA divided 14 vegetation indices into 4 categories, and the category which generally excluded the influences of soil and atmosphere (e.g. NDVI, SAVI and MSAVI) were the best. 3) Empirical dates showed that while the biomass was lower than 370 g/m², the yield estimation model was a simple unitary linear model, but when it was between 370 and 720 g/m², the simulation results of the linear and the exponential model were both very good. When the biomass was greater than 720 g/m², the exponential model was the best yield estimation model. Therefore, as the range of biomass increases, the yield estimation model should be gradually changed from a simple linear model to an exponential model.