The notion of bisimulation has been introduced as a powerful way to abstract from details of systems in the formal verification community. When applying to multiagent systems, classical bisimulations will allow one agent to make decisions based on full histories of others. Thus, as a general concept, classical bisimulations are unrealistically powerful for such systems. In this paper, we define a coarser notion of bisimulation under which an agent can only make realistic decisions based on information available to it. Our bisimulation still implies trace distribution equivalence of the systems, and , it allows a compositional abstraction framework of reasoning about the systems. Categories and Subject Descriptors