Abstract Since various types of distributed renewable energy resources are integrated into distribution systems in the form of microgrids, how to implement the coordinated operation between a distribution system and microgrids is a major concern. This paper focuses on solving the optimal power flow of a distribution system with multiple microgrids based on the decentralized saddle-point dynamics approach. First, the distribution system and the microgrids are regarded as separate entities and their external networks are replaced by Ward equivalent circuits. Hence, the distribution system and microgrids are decoupled so that their linearized power flow models can be built separately. Then, a decentralized quadratic programming model for optimal power flow with the distribution system and microgrids as separate entities is established to achieve low active power loss and high utilization of renewable energy resources. Next, the decentralized saddle-point dynamics approach is applied to solve this model from the viewpoint of the dynamic system control, which transforms the solution of Karsh-Kuhn-Tucker conditions into an asymptotically stable process. The proposed method only needs to exchange the border-bus voltages and equivalent injection powers between the distribution system and each microgrid, which can protect the privacy of different entities and possesses plug-and-play features. Finally, case studies on a real distribution system with two real microgrids are carried out to verify the effectiveness of the proposed method.