AbstractIn this article, a joint relay and channel selection problem is investigated in multi‐relay anti‐jamming communication system. Considering the jamming pattern and the relay node (RN) location distribution are unknown, the relay and channel selection problem is formulated as Markov decision processes (MDPs). Different from the existing research on anti‐jamming communication, in this article, the source node (SN) and all RNs are considered as agents who collaboratively learn the environment and make anti‐jamming decisions. A reinforcement‐learning‐based joint relay and channel selection method is proposed to achieve relay‐aided anti‐jamming communication. Specifically, the SN tries to make the optimal selection of relay which is under the least jamming threat, while the RN will access the jamming‐free channels. Various simulation results show that the algorithm can quickly learn the unknown changing pattern of the jamming environment, and make the effective joint relay and channel decisions to obtain high communication throughput which is close to the optimal.