With the development of economic globalization and the expansion of foreign trade, translation has been put forward with higher requirements. On this basis, a machine translation model based on deep learning is proposed. Traditional machine translation methods have many problems, such as rules, unstable statistical model, strong dependence on data, unable to deal with complex syntactic structure and so on. Aiming at the above problems, this project intends to build an automatic machine translation model based on multilayer neural network, and extract and learn features from it by using multilayer neural network. This method can improve the accuracy and fluency of the translation. The experimental results show that the fluency is improved to 89.4%. The purpose of this project is to explore the role of deep learning in automatic machine translation and to provide strong technical support for cross-lingual interaction and understanding.