Abstract: In order to meet the needs of operational weather forecast and research, a method of merging three sources of precipitation data, i.e. the rain gauge precipitation data collected at about 40, 000 automatic weather stations, the radar QPE (Quantity Precipitation Estimate) and the CMORPH satellite retrieved precipitation products, is developed to produce high resolution datasets at 0.05°×0.05° and 0.01°×0.01°, respectively. At 0.05° resolution, the biases of radar QPE and CMORPH products are corrected first based on the rain gauge data using the PDF (Probability Density Function) matching method. The bias of radar QPE is sharply reduced from -0.05 mm/h to -0.008 mm/h. The CMORPH is then merged with the radar QPE to produce an optimum first guess using the Bayesian Model Averaging (BMA) method. At 0.01° resolution, the 0.05° first guess is further downscaled (DS) to 0.01° using 1-km radar QPE data. The two first guesses are revised based on the rain gauge data using the Optimum interpolation (OI) method, and their errors are statistically sampled at the 0.01°×0.01° and 0.05°×0.05° resolutions, respectively. The verification at the 2, 419 independent stations shows that the two gauge-satellite-radar merged precipitation products at two resolutions are more accurate than any of the three sources of precipitation data and the gauge-satellite merged precipitation product. In particular, the data quality is well improved in areas of sparse gauge network in China. The three-source merged product at 0.05° resolution using the "PDF+BMA+OI" method is the best in general, while the one at 0.01° resolution using the "PDF+BMA+DS+OI" method has great advantages in detecting intense precipitation.