This paper introduces the overnight security service problem. The model obtained is a single-objective mixed integer programming problem. It is NP-hard in the strong sense, and exact approaches are not practicable when solving real-life instances. Thus, the model is solved heuristically, through a decomposition in two subproblem. The former is a capacitated clustering problem, the latter is a multiple travelling salesman problem with time windows. Both have a radius constraint which is unusual. The computational results prove the robustness of the approach used. Moreover, a detailed discussion of the results shows that the management objectives are satisfied, providing lower costs, a strong guarantee on the level of service and several different solutions.