The invention relates to a channel joint estimation method of a bidirectional multi-input multi-output relay system. The method has high channel estimation precision and low computation complexity, and the channel estimation can be quickly and accurately realized. The implementation steps of the method are as follows: 1) establishing a bidirectional multi-input multi-output relay system model; 2)designing a channel training signal; 3) encoding the received signal by a relay side and sending to a user; 4) constructing a TUCK-2 tensor model for the received signal by a user side; and 5) designing an iteration fitting algorithm with low complexity to realize the joint estimation of a channel matrix. The channel estimation method disclosed by the invention has the advantages of being high inprecision and low in computation complexity; furthermore, the channel can be effectively estimated even if the channel relevancy is enhanced.