Abstract: Distributed agent systems derive much of their power to solve complex, large-scale and open-ended problems because the agents within the system can communicate in highly flexible ways and patterns. However, before these systems can be widely deployed it is necessary to be able to precisely understand how different agent communication architectures perform in practice. This paper presents a methodology and performance evaluation study of agent communication architectures that use shared coordination spaces. We first propose a general performance model used to evaluate large systems of communicating agents. By defining several interaction styles our model can describe a wide variety of agent interaction patterns. We use this model to generate a synthetic workload measurement to evaluate the performance of an actual distributed agent system that uses a tuple-space type of coordination space. Our results show that the use of coordination space technology is a viable alternative for many types of communicating agent systems.