We study the merger paradox, a relative of Harsanyi's bargaining paradox, in an experiment. We examine bilateral mergers in experimental Cournot markets with initially three or four firms. Standard Cournot-Nash equilibrium predicts total outputs well. However, merged firms produce significantly more output than their competitors. As a result, mergers are not unprofitable. By analysing control treatments, we provide an explanation for these results based on the notion of aspiration levels, and show that the same logic also operates when a new firm enters a market. These results have some general consequences for adaptive play in changing environments. Copyright 2007 The Author(s). Journal compilation Royal Economic Society 2007.