Automatic gauge control of hot strip mills is realized via multi stands work together cooperatively, which is a typical multi-agent system, where the coupling effects among the automatic gauge control systems are strong. Duo to the parameter uncertainty and inevitably affected by external environment, it is hard to establish the accurate model of automatic gauge control system to improve the control performance. The paper considers a distribute control strategy of multi-agent automatic gauge control system of hot strip mills based on data driven control. The subspace identification technique is employed to formulate the distributed controller using the information of roll force and cylinder position. With the proposed control strategy, model identification processes is left out, and the effects of exit thickness deviation from upstream automatic gauge control systems can be reduced effectively, and both the cooperation of automatic gauge controllers and control precision are improved. Furthermore the sufficient condition for stability of the systems with the distribute control strategy is presented based on the proposed control strategy. Simulation results are used to verify the feasibility and effectiveness the proposed control strategy and corresponding results.