Modulation format recognition (MFR) has become a challenging topic with the development of multi-format hybrid transmission. Although MFR in communication systems has been widely studied, it is still not well developed for satellite communication systems. Deep learning (DL) has been demonstrated outstanding performance in quite a few fields. This paper studied the use of DL in the MFR for satellite communication systems. For research and application, DL relies on a large amount of data, which can be collected from communication systems. In addition, DL has the advantage of not requiring manual feature selection, which significantly decreases the complexity of MFR task. In this article, we used a DL model based on convolutional neural network (CNN) to test the accuracy of MFR. In addition, we also conducted comparative measurements with traditional ML-based algorithms. The simulation results showed remarkable performance of DL in MFR task of the satellite communication system, which proves the feasibility of its application.