In the present paper, MM5 model is used as the experimenting model. Wi th four kinds of cumulus parameterization schemes (AK、Grell、KF and BM), 48-hou r predicting experiments are made for 22 precipitation cases from Nov. 2001 to M ay 2002 under two kinds of model horizontal resolutions (45 and 25 km) and two k inds of PBL schemes (Blackadar and MRF). The results indicate the predicted rain fall error of KF or AK scheme is smaller, and the error of BM scheme is larger . Any scheme's predicting effect for one category precipitation is different from the effect for another category precipitation. When 24h rainfall is larger than 10 mm or 25 mm or 50 mm, no scheme's predicting effect is the best. The differen ce between different scheme's results for large category precipitation is larger than the difference for small category precipitation. Under most circumstances, same scheme's predicting effect under the resolution of 45 km is better than th at under the resolution of 25 km.