The net ecosystem productivity (NEP) is an important indicator to reflect the carbon budget of ecosystems. Our objective was to probe the NEP dynamic pattern and its response to climate change on the typical steppe in Inner Mongolia to better understand mechanisms of grassland carbon budget. BIOME-BGC model was used to simulate the NEP dynamics of four grassland communities in Xilingol River Basin by using the ecophysiological parameters according to the reference and the daily climate data simulated by MT-CLIM model as inputs. The water stress index (WSI),rain use efficiency (RUE) and carbon sequestration efficiency (CSE) were calculated based on other model outputs,including evapotranspiration (ET),net primary productivity (NPP) and gross primary productivity (GPP). During 1954 to 2012,the multi-year average NEP of Stipa baicalensis,S. grandis,S. krylovii and Leymus chinensis communities was 11. 41, - 7. 82,- 5. 03 and 9. 30 g C·m ~(-2)·a ~(-1),respectively. NEP seasonally exhibited a pattern of releasing CO_2 - sequestrating CO_2 - releasing CO_2 in a year. The order of multi-year average was S. baicalensis > L. chinensis > S. grandis > S. krylovii for WSI,S. baicalensis > S. krylovii > S. grandis > L. chinensis for RUE,and S. baicalensis > S. krylovii > S. grandis > L. chinensis for CSE. There was a close relationship between NEP and precipitation,and the correlations were at an extremely significant level in S. grandis,S. krylovii and L. chinensis communities. NEP turned from being negative to being positive when the annual precipitation of the four grassland communities was greater than 295. 76 mm.