A group of agents can be used to perform patrolling tasks in a variety of domains ranging from computer network administration to computer wargame simulations. Despite its wide range of potential applications, multiagent architectures for patrolling have not been studied in depth yet. First state of the art approaches used to deal with related problems cannot be easily adapted to the patrolling task specificity. Second, the existing patrollingspecific approaches are still in preliminary stages. In this paper, we present an original in-depth discussion of multi-agent patrolling task issues, as well as an empirical evaluation of possible solutions. In order to accomplish this study we have proposed different architectures of multi-agent systems, various evaluation criteria, two experimental scenarios, and we have implemented a patrolling simulator. The results show which kind of architecture can patrol an area more adequately according to the circumstances.