This paper proposes a novel distributed data-driven predictive control scheme to address the frequency control problem for Networked Microgrids (NMGs) in the presence of model uncertainty and disturbances. Firstly, the distributed data-based frequency model of NMGs is formulated according to input-output data. A suitable distributed data-driven controller is then proposed, and the convergence and stability of the system are analysed. A comparison with a model-based predictive control and with a state-of-the-art data-driven control methods is finally presented in simulation, showing the effectiveness and the ability of the proposed method to deal with a coupled frequency control problem without requiring accurate models of the system.