Abstract: North China is one of three major summer rainfall areas in eastern China. Precipitation over North China shows the characteristics of obvious emergency and locality. According to the statistics, 80%—90% precipitation occurs in June—August. Sometimes daily precipitation of a rainstorm can account for 50% precipitation amount of that month. Therefore, effective forecast is crucial especially for larger magnitude precipitation. Objective precipitation forecast is a difficult problem in NWP products interpretation at present. Objective precipitation forecast models are always established station by station, but larger magnitude precipitation is rare event for individual station. It is difficult to establish an effective forecast equation for an individual station. Precipitation intensity, spatial and temporal distribution over North China has its own particularity. Due to the regional characteristic, it is difficult to summarize in one model. Objective partitioning can be used in establishment of precipitation forecast model. Similar samples in the weather region are combined together. Regional forecast model is more stable than single-station forecast model, as the number of large-class precipitation samples increases.Seven weather divisions for summer precipitation over North China and adjacent areas are developed through Rotated Empirical Orthogonal Function (REOF) method, defined by the large contours of the seven REOF models. Objective precipitation forecast is based on probability regression precipitation categorical forecast. First, original precipitation is converted to 0 and 1 corresponding categories, and then forecast equations of different categories are developed to calculate each criterions. In real forecasting, the categorical precipitation will be determined through the criterion and the probability forecast of that category. Based on the daily precipitation data of station and T213 NWP products during the summer of 2006—2008, precipitation forecast model over North China and adjacent is established, which covers the domain (32°—42°N, 110°—124°E), including a total of 703 weather stations. Precipitation experiment is carried out for the summer of 2009 and 2010, and analysis of the forecast result indicates that regional method is better than single station method, especially for heavy precipitation. Regional model handles more factors than the single station model, so regional model makes a better prediction. Comparing to model direct forecast, regional forecast result is better, which reduces empty report obviously.