Abstract: In this paper, our objective is to facilitate the way in which emotion is conveyed through avatars in virtual environments. The established way of achieving this includes the end-user having to manually select his/her emotional state through a text base interface (using emoticons and/or keywords) and applying these pre-defined emotional states on avatars. In contrast to this rather trivial solution, we envisage a system that enables automatic extraction of emotion-related metadata from a video stream, most often originating from a webcam. Contrary to the seemingly trivial solution of sending entire video streams - which is an optimal solution but often prohibitive in terms of bandwidth usage - this metadata extraction process enables the system to be deployed in large-scale environments, as the bandwidth required for the communication channel is severely limited.