The derivation and applicability of reduced-order models for selection and assessment of plantwide control structures is studied. The paper demonstrates the advantage of exploiting the intrinsic structure of a chemical plant, which mirrors the decentralization of the control problem. The recommended procedure is to apply model reduction to individual units, followed by coupling these reduced-order models. This procedure is flexible and accurate, and leads to major reduction of the simulation time. The importance of retaining the model nonlinearity is highlighted.