Abstract: Input data aggregation is a common method for representing watersheds in the distributed-parameter models. To integrate a number of simpler process-based sub-models derived at different scale is one way to build distributed hydrological models. Both the spatial aggregation of watershed information and the model structure have significant impacts on the model output. This study investigated how different watershed partitioning affect flow,sediment and nutrient quantitatively using SWAT2000. The conclusions are:(1)Annual sediment yield and nutrients predictions are highly sensitive to the number of sub-basins used to represent a watershed,but runoff is not.(2)There're two thresholds for model output response. Below the lower threshold,although we can also get reasonable results by calibration,it might not fulfill the spatial demand for environmental management and programming. Above the higher threshold,distortional output may occur. The determination of the thresholds depends on the actual watershed characteristic. And(3)the sensitivity of sediment is bigger than TN and TP to the number of sub-watersheds when few subwatershed is partitioned,and smaller than TN and TP when more subwateshed is partitioned.