Abstract: The trailing phenomenon of α low energy spectrum is serious，and parameters meaning with the mathematical function fitting method are fuzzy.In view of these problems，it is proposed that the BP neural networks is applied to the α spectrum analysis.The model of BP neural networks is built based on MATLAB so as to predict the α spectral lines and estimate the element type respectively.Firstly，parameters that can represent the spectral information are selected as the input parameters for network training，and the α spectrum is predicted by using the nonlinear mapping function.Secondly，taking the full spectrum information as input data，the nuclide species are estimated by classifying the input data.According to the experimental results from comparison on the predicted spectrum and the original spectrum，its correlation coefficient is above 0.99，and the residual error range is around 2%.For the prediction results of nuclide types，the error for the prediction of two nuclides types is less than 1%.The research results above show that，BP neural networks method is accurate and simple，and is competent to do the α spectrum analysis work.

Key words: &alpha, energy spectrum, BP neural network, Predict, Classification