Self-optimizing control is achieved if a constant setpoint policy results in an acceptable (economic) loss L (without the need to reoptimize when disturbances occur). Skogestad (2000) presented a method for selection of controlled variables based on steady-state economics. The simplest is to select the setpoints of the controlled variables equal to their nominal optimum values. In this paper we extend the method by finding the robust optimal setpoints, or equivalently the optimal back-off from the nominal, for a given set of disturbances and implementation errors. As a case study we consider a reactor-separator-recycle process. For this process the control structures based on Luybens rule (“fix a flow in every recycle loop”) give infeasibility if we use the nominal optimal setpoints, but it is feasible with acceptable loss with robust optimal setpoints.