Abstract: Cognitive radio is one of the key technologies to alleviate shortage of spectrum resource. Medium Access Control (MAC) protocol recognition is of great practical value to extend the applications of cognitive radio. Traditional MAC recognition techniques suffer from low recognition accuracy and require manual feature extraction. In order to conquer these problems, this paper introduces the idea of deep learning and proposes a MAC protocol recognition method based on Long Short-term Memory (LSTM) network. The proposed method fully considers the temporal correlation of received signal, constructs a classifier using LSTM network, and realizes the recognition of four common MAC protocols. Experimental results show that, compared with traditional machine learning based algorithm, our method achieves higher recognition accuracy with less manual work and has great application potential.