The number of proposals for the National Natural Science Foundation is dramatically increasing in recent years. The prediction about the number is an important decision-making basis to macroscopically control the undesirable trend. This paper analyzes the application policies and the qualification factors, and then develops a method of predicting the number based on the potential applicants. The experimental result shows that the proposed method has much less average relative prediction errors than polynomial fitting, exponential smoothing, ARMA model and G(1,1) model. Moreover, the new method has an auxiliary capacity of analyzing the impact of the policies on the number of proposals.