Pré-publication, Document de travail: We study competition in the U.S. airline industry relaxing the Nash equilibrium assumption that airlines are able to predict perfectly the behavior of their competitors. We assess empirically whether an equilibrium is more likely to occur if it is the unique rationalizable outcome. We find that equilibria of short distance routes with high traffic and low concentration are the most fragile, and low-cost companies appear detrimental to their occurrence. Our analysis is applied to the measurement of welfare gains from firms’ entry, and to the characterization of the relevant market when some products are unobserved.