An efficient optimization procedure is proposed for computing a receding horizon control law for linear systems with linearly constrained control inputs and additive disturbances. The procedure uses an active set approach to solve the dynamic programming problem associated with the min-max optimization of an H-infinity performance index. The active constraint set is determined at each sampling instant using first-order necessary conditions for optimality. The computational complexity of each iteration of the algorithm depends linearly on the prediction horizon length. We discuss convergence, closed loop stability and bounds on the disturbance l(2)-gain in closed loop operation. (C) 2013 Elsevier Ltd. All rights reserved.