This paper considers receding horizon control strategies for constrained linear systems. Continuous-time predictions of input-output responses on an infinite horizon allow exploitation of the full range of plant behaviour. In addition, an infinite prediction horizon ensures the nominal stability of the receding horizon control law. The problem of optimizing predictions subject to constraints is made tractable through the use of a finite-dimensional prediction class and by minimizing prediction costs at discrete instants. Necessary and sufficient conditions on the prediction class are derived using input-output stability theory, and the controller structure further refined on the basis of a criterion for ensuring feasibility in a receding horizon sense. Constrained optimization is performed using quadratic programming.