Reconstruction of natural stream flow is fundamental to the sustainable management of water resources.In China,previous reconstructions from sparse and poor-quality gauge measurements have led to large biases in simulation of the interannual and seasonal variability of natural flows.Here we use a well-trained and tested land surface model coupled to a routing model with flow direction correction to reconstruct the first high-quality gauge-based natural streamflow dataset for China,covering all its 330 catchments during the period from 1961 to 2018.A stronger positive linear relationship holds between upstream routing cells and drainage areas,after flow direction correction to 330 catchments.We also introduce a parameter-uncertainty analysis framework including sensitivity analysis,optimiza-tion,and regionalization,which further minimizes biases between modeled and inferred natural stream-flow from natural or near-natural gauges.The resulting behavior of the natural hydrological system is represented properly by the model which achieves high skill metric values of the monthly streamflow,with about 83％ of the 330 catchments having Nash-Sutcliffe efficiency coefficient (NSE) ＞ 0.7,and about 56％ of the 330 catchments having Kling-Gupta efficiency coefficient (KGE) ＞ 0.7.The proposed construc-tion scheme has important implications for similar simulation studies in other regions,and the developed low bias long-term national datasets by statistical postprocessing should be useful in supporting river management activities in China.