[Objective] To estimate the net primary productivity (NPP) of grassland in China during the period from 1993 to 2015, and analyze the relationship between NPP and hydrothermal factors, in order to understand the productivity of grassland and the impacts of climate change on ecosystems.[Methods] Based on long-term remote sensing data, meteorological data and vegetation type data, we used CASA model (Carnegie-Ames-Stanford approach) to estimate the NPP of grassland in China. The variation of NPP and the responses of NPP to hydrothermal factors were calculated by using one-way linear regression, second-order partial correlation analysis and GIS spatial analysis.[Results] ① The total annual NPP of grassland was 7.60×10¹⁴ g (measured in term of carbon), and the annual average NPP per unit area was 296.76 g/(m²·a). The NPP of grassland decreased gradually from south to north, and from east to west in China. ② The total NPP of grassland in China decreased undulately with the linear rate of -1.415×10¹² g/a (p>0.05) from 1993 to 2015. The total NPP of grassland decreased significantly at a rate of -2.312×10¹² g/a (p<0.05) from 1993 to 2010, while increased at a rate of 7.00×10¹¹ g/a (p>0.05) from 2011 to 2015. In total, the area of grassland with decreasing NPP was larger than that with increasing NPP, and the NPP in 40.40% grassland changed significantly. ③ Annually, the second-order partial correlation between NPP and total solar radiation and precipitation was statistically significant, while that between NPP and annual mean temperature was insignificant. In terms of area, the significant partial correlation area between NPP and solar radiation was the biggest, followed by that between NPP and precipitation, and the significant partial correlation area between NPP and annual mean temperature was the smallest.[Conclusion] Grassland NPP in China declined widely while increased locally from 1993 to 2015. This variation was mainly affected by total solar radiation and precipitation.