Climate models predict, and empirical evidence confirms, that more extreme precipitation regimes are occurring in tandem with warmer atmospheric temperatures. These more extreme rainfall patterns are characterized by increased event size separated by longer within season drought periods and represent novel climatic conditions whose consequences for different ecosystem types are largely unknown. Here, we present results from an experiment in which more extreme rainfall patterns were imposed in three native grassland sites in the Central Plains Region of North America, USA. Along this 600 km precipitation–productivity gradient, there was strong sensitivity of temperate grasslands to more extreme growing season rainfall regimes, with responses of aboveground net primary productivity (ANPP) contingent on mean soil water levels for different grassland types. At the mesic end of the gradient (tallgrass prairie), longer dry intervals between events led to extended periods of below-average soil water content, increased plant water stress and reduced ANPP by 18%. The opposite response occurred at the dry end (semiarid steppe), where a shift to fewer, but larger, events increased periods of above-average soil water content, reduced seasonal plant water stress and resulted in a 30% increase in ANPP. At an intermediate mixed grass prairie site with high plant species richness, ANPP was most sensitive to more extreme rainfall regimes (70% increase). These results highlight the inherent complexity in predicting how terrestrial ecosystems will respond to forecast novel climate conditions as well as the difficulties in extending inferences from single site experiments across biomes. Even with no change in annual precipitation amount, ANPP responses in a relatively uniform physiographic region differed in both magnitude and direction in response to within season changes in rainfall event size/frequency.