This paper presents a decoupling algorithm for a large pressurized heavy water reactor to facilitate the design of a decentralized control system. The reactor models are generally high-order systems, which increases the difficulty of designing control systems. A convenient method of model reduction while maintaining the important dynamic characteristics of the process is through decoupling. The new decoupling algorithm proposed in this paper is used to create a decoupled system for decentralized controller design. To demonstrate the performance of this algorithm, a 72nd-order system was decoupled into three partitions, each containing 20, 27, and 25 states. Both a centralized controller based on the original model and decentralized controllers based on the decoupled model are designed. The advantage of the decentralized controller is shown through a fail-safe study.