Numerical experiments about the surface observation dat a assimilation in southwestern China were carried out by use of the GRAPES 3.0 as similation and prediction system. 0-48 h precipitation forecasts of a whole mont h and a heavy rainfall process that occurred in southwestern China in July, 2005 were analyzed. The main conclusions were as follows: Firstly, the influence of s urface observation data assimilation on precipitation forecast varies with the density of surface observation data used and the assimilating information select ed in this region; Secondly, when numerical forecast model takes higher resolution, the results of rainfall predic tion with all surface observation data, are superior to th ose with spare surface observation data from GTS; Thirdly, in complex terrain of so uthwestern China, assimilating wind information of surface observation data can have obvious influence on forecast results, decreases false precipitation ratio and m akes little contribution to raising missing precipitation and TS in contrast wit h not assimilating; Lastly, the scheme of assimilating relative humidity and geopotential height is the best one among these assimilation schemes.