In this paper a hybrid algorithm for the redundancy allocation problem is presented. The problem is the allocation of redundant components within series-parallel systems. We present an algorithm that deals with the classical formulation, where at least one component per subsystem must be included in the final configuration, as well as the k-out-of-n formulation, in which at least k components per subsystem must be included in the final network configuration. We propose a three-phase scheme in which the cross entropy method, the corridor method and a dynamic programming-based scheme are effectively intertwined. Computational results on well-known benchmark instances as well as on randomly generated large scale instances are presented, proving the effectiveness and robustness of the proposed algorithm.