Abstract: A better chaos sequence is generated by NN （neural network） through training the known nonlinear sequence. Then the sequence is used for mapping the plaintext to the cipher text. One advantage of the algorithm is that the hidden-mapping model of NN makes it difficult to get the direct mapping function of the ordinary chaos encryption algorithm. A better NN training method is achieved through theoretical analysis. Further more, the experiment shows that the output sequence has good random property as well as wonderful ability against decryption. From the practical point of view, the algorithm can be easily generalized to different ways by choosing the various known nonlinear sequences. The encryption speed of the algorithm is fast.