Hydrologic response in a watershed is driven by precipitation. Multi-site rainfall generators can be used to model watersheds using spatially varied rainfall inputs to better analyze how the rainfall variability affects runoff generation. This study adopted both a single-site rainfall generator (CLIGEN) and a multi site rainfall generator to generate two rainfall data sequences, which were then used to drive the Soil and Water Assessment Tool (SWAT) for runoff simulation. The 148-km(2) Walnut Gulch Experimental Watershed and its two sub-watersheds were selected to evaluate the hydrologic response. Runoff calibration was done against measured runoff in the watershed. Statistics showed that the single-site and multi-site rainfall generators gave similar results regarding annual precipitation. However, the multi-site generator performed much better than the single-site generator in both mean summer flow and for the different return period flows. The runoff derived from the single-site generator was significantly overestimated in all three watersheds. As for the multi-site generator, the derived runoff was satisfactorily predicted in the smaller watersheds but only overestimated in the largest watershed. This indicated that in small to medium sized watersheds, the spatial variability of rainfall could play an important role for hydrologic response because of the heterogeneity of convective rainfall in this semi-arid region, which makes the application of multi-site rainfall generator a better option than the single-site generator. (C) 2022 International Research and Training Center on Erosion and Sedimentation, China Water and Power Press, and China Institute of Water Resources and Hydropower Research. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd.