A BP neural network was established based on the following main experiment parameters of producing TbFe2 alloy by reduction-diffusion process: reaction temperature, holding time, quantity of Ca and particle size of Fe. A simulation was conducted, and the rate of conversion of TbFe2 alloy was predicted. The neural network was simulated and tested by 44 groups of experimental data. It can be concluded that the neural network has good performance to predict the rate of conversion of TbFe2 alloy. The design and the application of this neural network can help to shorten the periodic time of experiments, lower the experimental cost, and optimize the preparation processes.