An important task in the design of decentralized control systems for multivariable plants is the choice of the structure of interconnections between manipulated variables and controlled outputs, i.e. the control configuration. Most tools available for this task, such as the RGA, address mainly the stability properties of the overall system. In this paper we focus on performance, and consider in particular the problem of selecting control structures that enable a desired performance to be achieved through independent tuning of the subsystems. We show that, for this task, the common assumption of perfect control within the bandwidths of the subsystems is a poor one. Based on this, a new measure of interactions, the decentralized relative gain (dRG), is proposed. Finally, it is stressed that the effect of interactions on the magnitude as well as on the phase lag of the subsystems should be considered when selecting control configurations for performance.