We consider the problem of designing a scalable control architecture for large-scale variable-refrigerant-flow (VRF) systems. Using a gray-box modeling approach, and by exploiting the one-way coupling between the fluid dynamics and thermal dynamics, we derive individual linearized models for each class of dynamics. This sufficiently reduces the complexity of the problem so that a scalable analysis is possible. Based on the natural dynamics and coupling of the VRF system, which become apparent through the structure of the derived models, a partially decentralized control architecture is proposed. Communication is limited to one-way sensar information flow from the individual decentralized controllers to a global controller, leading to a simple yet highly effective control architecture which easily scales for systems with a large number of evaporators. The ability of the proposed control architecture and design to provide both high performance and reduced energy consumption is demonstrated through a simulated case study. (C) 2014 Elsevier Ltd. All rights reserved.