The fault diagnosis and prediction of satellites is a difficult problem due to the complex structure and the unique of operating environment of satellites as well as the presence of multi-source of satellite faults. Usually, one kind of reasoning model can only diagnose and predict one kind of satellite faults. This paper proposes a new method in which the use of multi-modal reasoning for satellite fault diagnosis and prediction is concerned. The method has been used in the development of the knowledge-based satellite fault diagnosis and recovery system and good results have been achieved.