In view of the subjectivity and “choose confliction” of the existing methods, a construction method for indicator system of financial risk early warning was proposed based on rough set theory. The new method can refine indicator system and identify the necessary index under reflecting comprehensively the risk condition by using the reduction of knowledge to eliminate redundant and noise of the sample data, and the process is objective and scientific which could avoid the “choose confliction” and the influence of subjective factors. The example result indicates that the mothed can reduce the complexity of the early-warning model and improve the prediction accuracy.