Aimed at the problems that slow convergence speed, poor learning performance and other imperfections exist in the classical BP neural network intrusion detection, a PCABP neural network intrusion detection method is put forward by adopting principal components analysis and additional momentum method, This method improves the classical BP neural network algorithm by data features selection and network weights amendment. Firstly, the paper standardizes the network data set, and then adopts it to deal with dimension reduction to confirm the characteristics. Finally, the paper detects the processed data set by improved BP neural network. Through the lots of experiments in KDD Cup 1999 network data sets, the result shows that the method has better performances in system model convergence, detection efficiency and detection accuracy in most network environment. Especially, in training samples, the convergence of system model, the detection efficiency and the detection accuracy are better than that by using BP neural network algorithm and halfsupervision intrusion detection algorithm.