Abstract: This paper focuses on the independent component analysis presenting a review on the basic model, the main method, the mathematical principle of blind source separation, analyzing the possibility of separation. The paper puts forward the method that based on the neural network unsupervised learning, also, improves the index on separation effects. Under the biology information processing background, combining artificial neural network with information theory to resolve this kind of problem can get the separation matrix adaptively by itself. It can separate mixtures which have both positive and negative kurtosis, and promotes the research of blind source separation greatly.