The problem of control-relevant experiment design is considered. We first develop the problem in abstract form as an optimization of an unspecified closed-loop performance index. Then, we make the problem specific by considering systems that can be described via a finite expansion of orthonormal bases and designing the controller by specifying the closed-loop system's sensitivity function. Although the resulting optimization problem is nonconvex and may admit local solutions, simulations of high-purity binary distillation columns are presented to demonstrate that significant improvements in controller performance may be obtained even by using the local solution.