Abstract: In order to make daily forecasting for the precipitation with MOS in the San-Hua area of Yellow River in the flood period of 15 July to 15 August,the experiments for selecting the predictors from a lot of variables of model output statistics of numerical forecast proposed,which are four models:stepwise regression of linear correlation,nonlinear stepwise egression,nonlinear stepwise regression with principal component and rotated principal component.The results show that the model of nonlinear stepwise regression with rotated principal component has the highest skill score of forecast in the models and is synoptic meaningful.