An effective coordination strategy between charging stations (CSs) and the distribution system operator (DSO) can fully exploit the charging flexibility of electric vehicles to improve the flexibility of power systems. Multiagent deep reinforcement learning (MADRL) has proven its effectiveness in coordination problems. However, existing MADRL-based approaches face the problems of incentive inequity and sample efficiency. To address these challenges, this article formulates the CSs-DSO coordination problem as a novel Stackelberg partially observable Markov decision process, where the involved agents, the DSO as leader and the CSs as followers, are rewarded according to their respective contributions. Furthermore, a knowledge-informed Stackelberg multiagent deep reinforcement learning (KI-SMADRL) strategy is proposed to solve this model, which integrates aggregation-allocation knowledge modules into CS agents’ learning loop to improve sample efficiency, and can coordinate CSs-DSO in a distributed manner. The effectiveness of the proposed strategy is verified by comparing its performance with multiple benchmarks.