This paper explores and evaluates the support for objectfocused collaboration provided by a desktop Collaborative Virtual Environment. The system was used to support an experimental ’design’ task. Video recordings of the participants’ activities facilitated an observational analysis of interaction in, and through, the virtual world. Observations include: problems due to fragmented views of embodiments in relation to shared objects; participants compensating with spoken accounts of their actions; and difficulties in understanding others’ perspectives. Design implications include: more explicit representations of actions than are provided by pseudo-humanoid embodiments; and navigation techniques that are sensitive to the actions of others. Keywords Social Interaction, Virtual Environments, Media Spaces, Object-focused Work.