The fraction of net primary productivity (NPP) allocated to belowground organs (f_BNPP) in grasslands is a critical parameter in global carbon cycle models; moreover, understanding the effect of precipitation changes on this parameter is vital to accurately estimating carbon sequestration in grassland ecosystems. However, how f_BNPP responds to temporal precipitation changes along a gradient from extreme drought to extreme wetness, remains unclear, mainly due to the lack of long-term data of belowground net primary productivity (BNPP) and the fact that most precipitation experiments did not have a gradient from extreme drought to extreme wetness. Here, by conducting both a precipitation gradient experiment (100–500 mm) and a long-term observational study (34 years) in the Inner Mongolia grassland, we showed that f_BNPP decreased linearly along the precipitation gradient from extreme drought to extreme wetness due to stronger responses in aboveground NPP to drought and wet conditions than those of BNPP. Our further meta-analysis in grasslands worldwide also indicated that f_BNPP increased when precipitation decreased, and the vice versa. Such a consistent pattern of f_BNPP response suggests that plants increase the belowground allocation with decreasing precipitation, while increase the aboveground allocation with increasing precipitation. Thus, the linearly decreasing response pattern in f_BNPP should be incorporated into models that forecast carbon sequestration in grassland ecosystems; failure to do so will lead to underestimation of the carbon stock in drought years and overestimation of the carbon stock in wet years in grasslands.