Recursive Least Square(RLS) is one of the main anti-jamming algorithms for adaptive array antennas. In order to improve the robustness of RLS and avoid the non-convergence caused by the improper forgetting factor, using the multi-channel received signal of adaptive array antenna, a new RLS algorithm based on its unbiased covariance matrix is proposed. Compared with the conventional RLS, the forgetting factor in this algorithm can accurately control the RLS iteration item. Consequently, the risk of algorithm non-convergence caused by the improper forgetting factor is reduced. The effectiveness is verified by the simulation