The modified Carnegie–Ames–Stanford Approach (CASA) model based on the comprehensive and sequential classification system of grasslands (CSCS, a unique vegetation classification system) was used to determine grassland net primary production (NPP) in Gansu province from 1982 to 2011 and its spatio-temporal variability. The relationship between NPP and climate drivers was analyzed. The results showed that annual NPP of grasslands in Gansu province averaged 139.30 gC m-2 yr -1 during the study period. NPP decreased from southeast to northwest across the province. Grassland NPP showed an increasing trend during the period 1982–2011, and the increase rate over the whole period was 92.91%. The highest NPP appeared in summer with more precipitation and higher cumulative temperature conditions; while the lowest values existed in winter. The largest correlation coefficient was found between the average annual NPP and the average annual precipitation (r = 0.77), followed by annual NPP and solar radiation (r = 0.70) or NDVI (r = 0.69), Annual NPP had no significant correlation with annual cumulative temperature (>0°C) or moisture index (K-value). Thus, precipitation is the major controlling factor on the average annual NPP in Gansu grassland. Solar radiation and NDVI also have important effects on grassland NPP in Gansu. These results may provide basic information for sustainable development and utilization of grassland and for the improvement and protection of the ecological environment as well.