In this paper, an optimal Smith-predictor design based on a generalized predictive control (GPC) approach is developed. The basic idea is to back-calculate the Smith control parameters based on the GPC control law, which has been designed to minimize a cost function. The design has turned out to be very simple to use. With a second-order model, the primary controller is simply a PID control. With a first-order model, it is a PI controller. The relationship between the control parameters obtained via the GPC approach and a pole placement approach is derived so that the latter approach may be used to derive initial GPC weights for further fine-tuning. Simulation and a real-time experiment illustrate the performance of the thus-designed optimal Smith control system.