Abstract: Based on the different assimilation schemes of cloud motion wind data, the effects of hourly data assimilation in GRAPES model on the prediction of a torrential rain process over low-mid Yangtze reaches on 11-12 July 2005 is investigated through the comparative analysis of the simulated results. Firstly, cloud motion winds at different levels were selected with three steps of quality control based on principles of temporal and spatial continuity, then, incorporated into hourly 3-D variational data assimilation. The initial fields assimilatedwith scheme 1 (using the quality-controlled cloud motion winds) and 2 (using one time level raw cloud motion winds) were compared with the NCEP reanalysis data, and impacts of cloud motion winds on the assimilated initial fields are discussed. At last, the 24 -hour precipitation of the torrential rain process on 11-12 July 2005 was predicted using the assimilated fields as the initial fields. Itwas found from analyses of the assimilated initial fields and predicted 24-hour precipitation fields, that three steps of quality control obviously reduced errors of the cloud motion vector data at various levels, assimilation of the quality controlled cloud motion winds into the initial fields improved the qualities of initial wind, geopotential height, and moisture fields, thus yielding the forecasts of more accurate location and intensity of the torrential rain.