Abstract: The precipitation is one of the most important meteorological factors that impact on human activities. The high resolution precipitation products are useful on weather and climate monitoring, calibration of numerical model and hydrologic prediction. Due to the asymmetry and discontinuity of precipitation in spatial and temporal distribution, the excessive and absent asymmetric rainfall resulted in the meteorological disasters such as flood and drought. In order to avoid and reduce these disasters, it is important to know the amount and distribution of precipitation exactly. Using the rainfall recorder observation is the most direct, effective and common method to obtain precipitation information. But the rainfall recorder data only represent several isolated stations information, not the continuous grid data. In order to obtain the standard grid data, it is necessary to process the original rainfall data with math methods. A lot of interpolation methods have been proposed, such as Cressman, Barnes, Gandin, Shepard, Hulme. Bussieres and Hogg as well as Chen have compared and verified the interpolation methods in application of the daily and monthly rainfall analysis. It has been concluded that Gandin optimal interpolation possesses much more exactly, effectively and stably. But for smaller time scales, what math interpolation can be used to obtain a better results and how to set the correlative parameters needs further researches. Using the Parameter elevation Regressions on Slope Model (PRISM), an objective analysis of gauge hourly regional observation precipitation in Guangdong and Guangxi from 20 May to 30 August 2007 with the spatial resolution of 0.125°×0.125° by Shepard interpolation method are implemented, the results are compared with optimum interpolation method too. The comparison analysis shows that both the results can represent the inter seasonal and diurnal variation of gauge regional mean precipitation well between direct interpolation and PRISM interpolation. Although the spatial distribution of precipitation gained by the direct interpolation is smoother than that by PRISM interpolation, but the estimated precision of direct interpolation is lower than that of PRISM interpolation. The cross validation indicates that the accuracy of precipitation products by the optimum interpolation is better than that by Shepard interpolation. The PRISM optimum interpolation is an optimal method that can improve correlation and reduce root mean square error as well as system error.