In the existing MIMOOFDM relay system, most channel estimation using the supervised channel estimation algorithm with training sequence has lower spectrum utilization. In this paper, based on the MIMOOFDM relay system and signal modeling, a channel estimation of MIMOOFDM relay system based on embedded PARAFAC model is proposed. Firstly, the KhatriRao decomposition algorithm is used to estimate the relaythe destinationend matrix, and then use the alternating least squares algorithm to estimate the sourcerelay channel matrix and the source matrix. Finally, the computational complexity and uniqueness of the algorithm are analyzed. The simulation results show that the algorithm has lower computational complexity and faster convergence rate than the trilinear least squares algorithm, and the performance is close to the traditional supervised channel estimation algorithm with training sequence.