One of the most desirable characteristics of an intelligent interactive system is its capability of interacting with users in a natural way. An example of such a system is the embodied conversational agent (ECA) that has a humanoid aspect and the capability of communicating with users through multiple modalities such as voice, gesture, facial expressions, that are typical of human-human communication. It is important to make an ECA able to fit well in each role in a conversation: the agent should behave in a realistic and human-like way both while speaking and listening. So far most of the work on ECAs have focused on the importance of the ECA's behaviour in the role of the speaker, implementing models for the generation of verbal and non-verbal signals; but currently we are mainly interested in modelling the listening behaviour. In this paper we will describe our work in progress on this matter.