For the problem of automatic flatness control (AFC) in cold tandem mills this paper proposes control techniques based on quadratic optimization and delay compensation. Three different strategies are presented and compared: a centralized solution based on a global quadratic programming (QP) problem that decides the commands to all the actuators, and two decentralized solutions where each actuator command is optimized locally. All schemes are based on a global exchange of information about the commands generated at the previous time step at each stand to compensate for the numerous delays present in the mill. Control algorithms are tested in simulation considering a tandem mill with five stands as a benchmark, and results are shown to demonstrate the performance of the proposed schemes. (C) 2010 Elsevier Ltd. All rights reserved.