Software agents represent an interesting paradigm to develop intelligent and distributed systems, because of their autonomy, proactiveness and reactivity; in addition, their sociality enables the distribution of the application logic in different agents that can interact together and with the host environment. In such scenario interactions must be carefully designed and managed at run-time. The concept of role has been adopted in different (agent) approaches to flexibly manage interactions; roles represent stereotypes of behavior, which are useful not only in the modeling of systems and applications, but also in their design and implementation. Overhearing is a technique that enables the observation of agents� behavior by �hearing� the exchanged messages. In general, overhearing can be useful to �label� observed agents and manage agent organizations. In this paper we explore the adoption of overhearing in conjunction with agent roles in order to provide more features to agents themselves. In particular the approach presented here can support and help the agent deciding the role to assume and then how to use it, after having observed the agent behavior.