As the role of virtual teams in organizations becomes increasingly important, it is crucial that companies identify and leverage team members’ knowledge. Yet, little is known of how virtual team members come to recognize one another’s knowledge, trust one another’s expertise, and coordinate their knowledge effectively. In this study, we develop a model of how three behavioral dimensions associated with transactive memory systems (TMS) in virtual teams—expertise location, task–knowledge coordination, and cognition-based trust—and their impacts on team performance change over time. Drawing on the data from a study that involves 38 virtual teams of MBA students performing a complex web-based business simulation game over an 8-week period, we found that in the early stage of the project, the frequency and volume of task-oriented communications among team members played an important role in forming expertise location and cognition-based trust. Once TMS were established, however, task-oriented communication became less important. Instead, toward the end of the project, task–knowledge coordination emerges as a key construct that influences team performance, mediating the impact of all other constructs. Our study demonstrates that TMS can be formed even in virtual team environments where interactions take place solely through electronic media, although they take a relatively long time to develop. Furthermore, our findings show that, once developed, TMS become essential to performing tasks effectively in virtual teams.