This paper describes two methods for analyzing interactions in collaborative virtual environments (CVEs): one whereby quantitative data are captured, interaction is categorized into a number of activities, and statistical analysis can be performed on frequencies and sequences of events. The other is based on the transcription of individual fragments of interaction, which are analyzed in terms of their key dynamics. The two methods each have their strengths and weaknesses, especially in terms of generalizability and the lessons we can derive from them. Both also point to different problems that need to be addressed in methods for analyzing interaction—such analysis being, in turn, a precondition for improving the usability of CVEs. The paper concludes with an argument for a combination of the two methods, and some reflections about the relationship between the analysis of interaction and the usability of CVEs.