Because of the shortage of the traditional methods for evaluating regional innovation capability, it presents a new method based on rough set and BP neural network in this paper. Firstly, index system for evaluating about regional innovation capability is set up, and then rough set is used to simplify the attributes of the index system, it will ease the burden of training and learning of BP network and reduce the amount of computing. The simplified attributes are input into the network which has been designed. Through the training and testing of lots of studied samples, the error of the model is limited in a preconcert range. Thus, an integrative network model has been established. At the end, it proves the accuracy and feasibility of the method with an example.