Ecologists are being challenged to predict ecosystem responses under changing climatic conditions. Although air temperatures are increasing, the magnitude and direction of change in precipitation (increase or decrease) are uncertain for many sites. Given that water availability is the primary driver of ecosystem processes in temperate grasslands and shrublands, these uncertainties in precipitation mean that ecologists must account for both possibilities. Long-term research sites (LTER, LTAR) provide natural experiments for ecological responses that occurred historically during multi-year drought or wet periods that can be used to make predictions under future climate scenarios. We tested three alternative hypotheses using long-term data (12 to > 50y) of aboveground net primary production (ANPP) from seven sites in North America where precipitation showed sequences (> 3 years) of wet periods, multi-year drought, and no trend years. We hypothesized that ANPP in wet (or drought) periods can be best explained by: (1) long-term relationships between ANPP and precipitation, (2) relationships between ANPP and precipitation in individual wet or dry years, or (3) relationships between ANPP and precipitation in wet or dry periods. We compared r2 values among equations at each site to determine the relationship with the best fit. For most sites across the region, the equation developed using ANPP and precipitation during drought periods was a better predictor of ANPP during drought compared with the long-term equation or the equation using individual dry years. In addition, the drought period equation had a steeper slope than the long-term equation. Thus, approaches that use long-term ANPP-precipitation relationships to predict ANPP during multi-year drought will result in over-estimates of ANPP. In contrast, in wet periods at some sites, the number of wet years in a row was a better predictor of ANPP than the amount of rainfall during the wet period. Cumulative processes, including plant-soil water feedbacks, sequential plant population processes, and plant or soil legacies may be operating to influence these temporal dynamics. These equations relating ANPP to precipitation during drought or number of wet years can be used to explain historic patterns, such as the 1930s drought or unusual grass recovery patterns, as well as improve future predictions under directional climate change.