The rainfall generation and confluence in small mountainous watersheds exhibit short and rapid runoff response, characterized by abrupt rises and falls during flood events. The short duration and complexity of flood processes pose challenges for flood forecasting in these regions. Moreover, some remote small mountainous watersheds lack well-equipped monitoring stations, resulting in limited hydrological data availability. These areas are considered to appertain lacking or without data regions, where traditional lumped or distributed hydrological models face limitations. Therefore, it is of great significance to conduct the research on the runoff generation and confluence model in small mountainous watersheds. The improved SCS-CN runoff model and the general unit hydrograph confluence method were coupled, and a case at the upper reaches of the Jiuzhou hydrological station in the Xizhijiang River basin, Pearl River basin, was selected for the validation of the proposed model. The three sources Xin'anjiang model was chosen as a reference model for comparison. The flood relative error (Evol), peak relative error (Epeak), peak occurrence time error (Tlag), and Nash-Sutcliffe efficiency coefficient (Dc) as evaluation indicators for model performance. The applicability of the established runoff model in flood simulation in small mountainous watersheds was examined. Results showed that both models demonstrated the applicability in flood simulation in the study basin, with the Xin'anjiang model exhibiting higher accuracy than the SCS-CN general unit hydrograph model. For all 13 flood events, the Xin'anjiang model produced the average Dc was 0.79, with 10 events meeting all accuracy requirements (Evol |≤20%, |Epeak|≤20%, |Tlag|≤3h,and Dc≥0.6), achieving a pass rate of 77%. Regarding the proposed SCS-CN general unit hydrograph model, the average Dc was 0.70, with 9 events meeting all accuracy requirements, achieving a pass rate of 69%. Overall, the accuracy of the Xin'anjiang model was better than the proposed model in the study basin. However, considering of the newly proposed SCS-CN general unit hydrograph model featured a straightforward structure, with parameters that were easy to determine, and was capable of yielding results with acceptable accuracy. It presented a new modeling option for flood forecasting in small mountainous watersheds. Simultaneously, it also provided support for the research of the regionalization law of parameters in general unit hydrograph model.