The chaotic neural networks with the characteristics of very complicated nonlinear dynamics have been a hot project of application to secure communications in recent years. This paper firstly deals with some main theoretic models, the characteristics and complexities of nonlinear dynamics of them. It analyzes the basic principles which chaotic neural networks are applied to secure communications using the complex property of nonlinear dynamics of them such as chaotic synchronization and chaotic attractors. It finally summarize the projects of chaotic neural network and its application to secure communications that needs to be further researched.