Much attention has been paid recently to controltheoretic research on what is called feedback error learning (FEL). This paper provides yet another FEL control scheme, which generalizes existing ones in two points: i) It can treat a plant which is the product of known and unknown transfer matrices, and thereby reduces the number of parameters to be learned. ii) The response model is less restrictive, due to a general two-degree-of-freedom structure. Numerical simulation illustrates the effectiveness of the proposed scheme.