Aimed at the uncertainty of safety risk prediction in construction project currently, Rough set was used to pretreat the safety factors of construction project, and then the pretreatment results were put、the information structure of support by cetor macher(SVM)to traing and prediction. So the construction project safety risk early- warning model was estaldished, and the validity of the orithm was proved by empirical analysis. The MSE of SVM is 0. 011 5，by the BP neural network is 0. 070 7. Fxperimental re- sups revealed that the model of rough set一support vector machine(R.S一SVNI) can realize to delete the superfluous inputting information, to reduce the complexity and improve the interfere resistance. In small sample conditions, the early一warning model of RS- SVM is greatly superior to BP neural network in pre- diction precision and generalization ability.