When a plant-wide control structure is being designed, there are often more measurements and manipulations than are required. Therefore the designer must judiciously select a subset of them that will be used for feedback control. If the control structure is to be decentralized, a set of pairings between the selected measurements and manipulations must also be established. This subset of process variables and their feedback interconnections has been termed the control configuration, and must be decided prior to the design of the controller. However the number of possible control configurations is usually very high, and the complexity of the controller depends upon the control configuration chosen. This paper presents a systematic way to design the control configuration so as to reduce controller complexity, and yet meet control objectives in the presence of uncertainty. This method, called control configuration design, is then applied to a plant-wide central problem posed by Tennessee Eastman.