Theoretical design of fuel has always been the focus of research about fuel in the area of propulsion technology. It can effectively overcome the complexity and potential danger of the experiment, and guide experimental synthesis of fuel, which can be verified by experimental results. It is anticipated that a new generation of fuel can be efficiently designed for subsequent fuel synthesis and application. However, the traditional theoretical calculation methods, such as group contribution method and quantum chemical method, have the defects of low accuracy and efficiency. Machine learning, a rapidly developed algorithm, has opened up a new way to design potential high-energy fuels, which exhibits strong capabilities in both property prediction and molecule design. In this review, several fuel molecule descriptors for machine learning are introduced, and different machine learning models for fuel property prediction and molecule design are briefed. Furthermore, the research on machine learning assisted property prediction and new molecule design of fuel are summarized respectively. Finally, the challenges and future development of machine learning applied in fuel design are discussed.