Abstract: Flood disaster is one of the important factors that restrict the sustainable development of the economy and society of China. The development of a well-performing rainstorm-flood forecasting system is an important non-engineering flood prevention measure that would mitigate the loss of imminent flood disasters. A rainstorm-caused flood forecasting system, which is based on the distributed hydrological model CREST V2.1, is developed to provide refined streamflow, evapotranspiration, soil moisture, and other forecast products. By utilizing operational precipitation data from China Meteorological Administration (CMA) to serve as input for this system, nationwide flood forecasting is carried out by 0.125°×0.125° daily, and regional forecast is done by 30"×30" hourly. For the former, one typical watershed is selected for each of ten river basins as Songhua, Liao, Hai, Yellow, Huai, Yangtze, Southeast, Pearl, Southwest and Northwest River Basins, while for the latter just the Huai River Basin is taken as focus. The SCE-UA optimization algorithm is adopted to search the optimal parameter sets that maximize the Nash-Sutcliffe efficiency (E) between the observed and the simulated streamflow discharges for gauging stations of typical watersheds. E, correlation coefficient (C), and relative bias (B) are used to evaluate model performances before and after the calibration of model parameters. Validation tests are conducted by transferring calibrated parameter values to another flood event of the same watershed. Results show that the calibrated model can reproduce the observed flood processes and provide accurate hydrological forecasting service. Compared to the non-calibrated model, the calibrated one significantly improves E and B, and moderately improves C. It has good applicability in watersheds with different hydroclimatic, geological and geomorphological conditions, but has relatively weak forecasting ability for frequently fluctuating low-flow flood. For the model parameters, their values not only depend on the hydroclimatic, soil and vegetation conditions of the watersheds, but are also influenced by interactions among physical processes of the model. Besides, some empirical parameters need to be calibrated according to different levels of the flood events for the same watershed. Generally, this flood forecasting system show good forecasting accuracy and timeliness, which meets operational needs. However, further work is still needed to improve the prediction accuracy of the model. For example, the snowmelt module could be implemented into the CREST model to improve the prediction accuracy for flood disasters caused by snowmelt in the Northwest, Northeast, and Qinghai-Tibet Plateau regions. In addition, more observed streamflow discharge data should be collected to help calibrating model parameters for more watersheds. Furthermore, uncertainty quantification methods should be adopted to understand parameter behaviors, quantify and reduce parametric uncertainties.