Exploring the watershed subdivision threshold in Soil and Water Assessment Tool model is crucial for accurate hillslope soil erosion estimation, which has been rarely explored. This study investigated impacts of watershed partitioning on simulation of hillslope sediment generation and its spatial variations, and determined a partition threshold using three newly proposed evaluation indicators (coefficient of spatial variations for hillslope sediment generation, change rate of hillslope sediment generation, and change rate of coefficient of spatial variations for hillslope sediment generation). The Xixi basin in the south-eastern coastal region of China was taken as an example, and was partitioned into eleven different watershed delineations. Effects of watershed subdivision on model parameter aggregation and areal distribution of hydrological response units were analyzed. The results indicated that (1) the previous method revealing the effects of watershed subdivision on hillslope sediment generation, which was only based on watershed outlet data, and the partition threshold determined by the method, were unreliable. (2) Hillslope sediment generation was seriously affected by watershed subdivision levels, increasing the number of sub-watersheds would decrease the modelled amount of hillslope sediment generation and increase its spatial variations. The partition threshold for the average sub-watershed area was 1.6% of the total watershed area and the number of sub-watersheds was 61 as determined from our new indicators, which resulted in stable amount of hillslope sediment generation and its spatial distribution, thus, the hillslope sediment generation could be adequately simulated by this method. (3) The hydrological response unit area, which was strongly related to the change in hillslope sediment generation, was very sensitive to watershed subdivision until the level of watershed subdivision approached the threshold.