<p>The cooperative output regulation problem for unknown linear multi-agent systems has been studied by both policy-iteration method and value-iteration method via distributed internal model approach. However, the original results were limited to single-input single-output linear multi-agent systems under the assumption that the communication digraph is acyclic. Recently, the authors have extended the existing result to multi-input multi-output linear multi-agent systems over a general static and connected digraph by a more efficient value-iteration method. Since the policy-iteration method is simpler and has a much faster convergence rate than the value-iteration method, in this paper, the authors further apply the policy-iteration method to the cooperative output regulation problem of unknown multi-input multi-output multi-agent systems over a general static and connected digraph. Compared with the existing policy-iteration method, the proposed policy-iteration approach not only drastically reduces the computational cost, but also significantly weakens the solvability conditions. Moreover, by introducing a virtual exosystem, the proposed policy-iteration approach eliminates the need for employing a distributed observer. As a result, the data collection can start at any time, and the computing cost for each agent is also reduced.</p>