A novel decomposed optimization method based on price response function is proposed for solving security constrained economic dispatch problem over large-scale power systems.In this method，provincial power systems are taken as basic optimization units，and the global optimization problem is decomposed into an upper-level master problem and provincial sub-problems.In the master problem，network reduction technique is imposed，and the dynamic price response function is trained based on the results of the sub-problems in order to optimize the power flow of inter-province tie-lines.In the sub-problems，power output of generating units is optimized with fixed tie-line export，with marginal pricing information corresponding to power loads of different time periods calculated and submitted to the upper-level.The proposed method fully exploits and utilizes the implicit pricing information in the sub-problem results of each iteration，thereby avoiding the oscillation from which the conventional method of Lagrange relaxation suffers much because of the monotony of pricing signal.Hence，the method is able to converge quickly in a few steps with good robustness and sufficient accuracy from the engineering point of view.Computational results of IEEE RTS and a large-scale real power system are presented to verify the effectiveness of the method.