To solve the problem that the simulated river network extracted directly from the original DEM is not accurate, deviates from the actual water system, and affects the sub-basin division. This paper based on the actual imagery of the region, firstly, lowers the elevation of the original DEM through the actual river positions in the regional actual imagery; secondly, determines the optimal watershed area threshold of the river source by the river network density avoiding subjective threshold selection; thirdly, sets different sub-basin watershed area thresholds, and determines the appropriate sub-basin watershed area threshold and the number of sub-basins through the relationship between the threshold and the number of sub-basins; finally, by comparing the hydrological process simulation results of the three different sub-basin division schemes, the scientificity of the sub-basin division is verified. Taoxi watershed is located in Shucheng County, Lu "an City, Anhui Province.The application study in Taoxi River Basin shows that: in GIS, the elevation of the DEM grid passing through the actual water system is lowered by 40 m firstly, and then the optimal watershed area threshold determined by the river network density avoiding subjective threshold selection is 0.81 km2, the extracted river network is closer to the actual; the number of sub-basins in Taoxi River Basin is divided into 7 by setting the threshold interval of 108~144 km2, the average area is 213.19 km2, the sub-basin area division is consistent with the standard of the first-class small watershed range in China (50~300 km2). Comparing the distributional simulation of Taoxi River basin under three schemes A, B, and C, which were divided into 7 sub-basins and 11 sub-basins after the DEM grid elevation was lowered by 40 m along the main river channel and the original DEM, respectively, and the model parameters were calibrated using a genetic algorithm, the results show that: in daily discharge simulation, with 1982—2012 as the training sample and 2013—2022 as the validation sample, the deterministic coefficient of the training sample discharge simulation was 0.87, 0.87, and 0.86, and the validation sample was 0.9, 0.91, and 0.9, and the simulation accuracy of the three schemes was basically consistent; selecting 20 maximum flood processes within the period of 1982—2021 for simulation, peak discharge, discharge volume, and peak occurrence time were used as accuracy evaluation indicators, and the qualified occasions of schemes A, B, and C were 19, 19, and 16, respectively, with qualified rates of 95%, 95%, and 80%; the average deterministic coefficient of the 20 flood process simulations for schemes A, B, and C were 0.93, 0.94, and 0.93, respectively, and the accuracy evaluation indicators of the first and second grades for schemes A were 16 and 4, respectively; the first and second grades for schemes B were 18 and 2, respectively; and the first and second grades for schemes C were 18 and 2, respectively. In summary, according to the actual river system, dividing the river basin into 7 sub-basins after lowering the elevation of the DEM grid along the main river channel can improve the qualified rate and the accuracy of the simulation, and dividing the Taoxi River basin into 7 sub-basins according to the area range of the first-order sub-basin is similar in terms of the deterministic coefficient and other accuracy indicators compared with dividing it into 11 sub-basins. The research method can provide reference for the construction of distributed hydrological models and the division of sub-basins in similar conditions.