Unit restarting after a blackout is the basis of the overall restoration process for a power system. A reasonable restarting sequence of units may speed up the restoration process and reduce economic loss. In view of the situation that there are usually several black-start sources in a large-scale power system, a global optimisation method for unit restarting considering black-start zone partitioning is proposed in this study. On the basis of the multiple time-step modelling framework, the optimisation problem is formulated as a mixed integer linear programming model, with the objective functions to maximise the total-weighted MWh output and minimise the links between the subsystems under various operational constraints. The hierarchical method based on generalised Benders decomposition is applied to decompose the optimisation problem into one master problem and two sub-problems: (i) determination of the unit restarting sequence considering only the grouping of units and (ii) rationality check of unit grouping and check of operational constraints based on power flow calculation. The master problem and sub-problems are alternately and iteratively solved until the optimal solution is obtained. The results of New England 10-unit 39-bus system and a practical power system in southwest China verify the validity of the proposed method.