Several hypotheses derived from an information sampling model of group discussion were tested with 3-person teams of physicians given 2 hypothetical medical cases to diagnose. Some of the information about each case was given to all 3 team members before discussion (shared information), whereas the rest was divided among them (unshared information). As predicted, shared information was, overall, more likely to be discussed than unshared information, and it was brought into discussion earlier. In addition, it was found that team leaders repeated substantially more case information than did other members and that, over time, they repeated unshared information at a steadily increasing rate. The latter findings are interpreted as evidence of leaders' information management role in problem-solving discussions.