This paper presents a new method for restoration of distribution networks after a fault occurrence. This problem is solved from the viewpoint of the distribution system operator with the main goal of minimizing the operating cost during the fault clearance period. The effects of distributed generation (DG) units and direct load control (DLC) programs are considered in designing the proposed restoration procedure. Moreover, the uncertainties associated with the predicted loads of different nodes and the availability of DG are modeled here. Robust optimization is used to model the uncertainties of restoration problems and manage their associated risks. Finally, a robust reconfiguration plan is obtained solving a bilevel problem using a genetic algorithm (GA). The upper-level problem is concerned with evaluating the optimum configuration by GA and the lower-level problem obtains the optimum schedules of DG and DLC with an AC optimal power flow. A 32-bus test system is used to demonstrate the applicability of the proposed method.