Grassland productivity is the most direct indicator of grassland’s ecological status; thus, precise and rapid estimation of grassland productivity has an important significance for evaluating regional ecological carrying capacity, utilizing and developing natural resources reasonably. In this study, based on Moderate Resolution Imaging Spectroradiometer remote-sensing data, meteorological data, and ground measured data, the light utilization efficiency parameter of Carnegie–Ames–Stanford Approach model was optimized, and the optimized model was used to estimate net primary productivity (NPP) and analyse its spatio-temporal distribution of Xilingol grassland between 2005 and 2014. In addition, we investigated the NPP response mechanism to climate factors from two aspects, i.e. the accumulation amount and time of NPP. The results indicated that (i) Maximum light utilization efficiency was calculated to be 0.539 g C MJ⁻¹ for Xilingol grassland according to the least error criterion. (ii) Validation against NPP observations from 30 sites showed good performance of the optimized model in the research area, with an overall coefficient of determination (R²) of 0.72 and mean relative estimation error of 0.29. (iii) The annual mean NPP of Xilingol grassland was 161.23 g C m^–2 year^–1, with a decreasing trend from the northeast to the southwest. (iv) NPP presented increasing trend with a rate of 3.89 g C m^–2 year^–1 overall during 20052014, and the largest increase in NPP was found in temperate meadow-steppe. However, desert ecosystems showed the largest increasing rate after NPP standardization. (v) Precipitation was the main factor for driving carbon accumulation at the inter-annual scale, and the time to maximum NPP was significantly correlated with spring precipitation (R² = 0.637), NPP peaked early as the spring precipitation increasing.