A method which combines the rough set and ANN is proposed, and it is applied to the financial risk research of Chinese listed companies. By empiric analyzing on the financial date of Chinese listed companies, the results show that the introduction of the rough set cuts down the input dimension of ANN, and the ANN algorithm is improved by the momentum accession and the parameter self-adaptation. The modified algorithm has advantages over the traditional BP ANN in the veracity and precision of network training.