Grassland ecosystem is one of the most widely distributed types in the terrestrial ecosystems. Estimating carbon storage in grassland ecosystem has been a central focus of global change researches. In order to estimate the grassland net primary productivity (NPP) quickly and reliably, based on the field survey data and the remote sensing image data of the same period, the comprehensive estimation model of grassland NPP in China was developed by using normalized difference vegetation index (NDVI) and climate data. According to the basic principles of grassland genetic and the relationship between the single factor and the NPP, through statistical analysis, the model structural factors were put forward, and then integrated together. The comprehensive model included two sub models of leaf area index (LAI) and photosynthetic accumulation (PA), and it was NPP= LAI×PA. The remote sensing data NDVI was used as a driving factor for constructing LAI sub model and it was LAI=ln(5.79×NDVI+5.91)/(2.73-2.46×NDVI). The climate data such as temperature, precipitation, and radiation were used as driving factors to construct PA sub model. In the PA sub model, there was a logarithmic relationship between the grassland NPP and mean monthly temperature, and the correlation coefficient r=0.4382 (P<0.01, n=95). There was a linear positive correlation between the grassland NPP and monthly rainfall, and the correlation coefficient r=0.6626 (P<0.01, n=95). There was an exponential relationship between the grassland NPP and radiation, and the correlation coefficient r=-0.7047 (P<0.01, n=95). So PA sub model was described as PA=ln(2+T/18.1)×Sqrt(W/89.3)×110/Exp(R/603-0.8), where T was mean monthly temperature, W was monthly rainfall, and R was monthly radiation. The model was validated by independent measured data which was not used for constructing the model. There was a good correlation between the simulated and observed NPP values, and the R² was 0.8519 (P<0.01). The root mean square error (RMSE) and the relative root mean square error (RRMSE) were 59.955 gC/m² and 0.358, respectively. The small values of RMSE and RRMSE indicated that the model was reliable. The average relative error between the simulated and measured values was only 1.97%, and the model can accurately predict NPP. So it was feasible to estimate grassland NPP in China by using this model, and this model provided a new method for estimating of grassland NPP in China.