Abstract: A new method for evaluating engineering project based on the rough neural network was proposed. This method applying the rough set theory carries on the pretreatment to the input data of nerve network, and withdraws essential ingredient as the input of the network. It can improve the convergence ratio and the approaching precision. Taking China agricultural water engineering project as the case, the results indicate that the model is not only more fair but also reasonable. It improved the learning efficiency of the neural network model.