Abstract: Accurate simulation of rainfall-runoff is important in urban flood prevention and mitigation．An important component of urban flood is baseflow-basic runoff that comes from groundwater aquifer and flows in river channel．To improve simulation accuracy, a storm water management model (SWMM) was developed．Discharge data required for model calibration was processed by constant-discharge baseflow separation．Discharge data with or without baseflow separation was used for model calibration and validation respectively, to simulate six different historical floods, Effect of baseflow separation on simulation results was quantitatively assessed．Baseflow was found to account for 10% of peak discharge, but with significant impact on flood process simulation．When original discharge data was used for model calibration, and discharge from precipitation and other sources (e.g. groundwater) were not distinguished, overall model performance was not satisfied with average Nash-Sutcliffe coefficient of efficiency (C_NSE) of 0.532. Simulation accuracy was improved significantly when baseflow was separated from original discharge, with an average C_NSE of 0.765．After adopting baseflow separation, simulation accuracy (C_NSE) for heavy rainstorms increased on average by 43.7%, relative error of peak discharge decreased by 3.59%．Baseflow, however, had no obvious influence on flood peak time．The present work provides some scientific basis for urban rainfall-runoff and flooding simulations, and would be helpful for water authorities to formulate their disaster prevention and mitigation measures．These data expand the application of baseflow separation in urban storm flood process simulation, and provide scientific basis and technical support in urban storm flood disaster prediction and early warning．