Abstract This article investigates whether ‘rational expectations’ are really rational, i.e. whether they maximize agents' utilities in some way. For that purpose we construct a model where each agent may have a whole array of expectations schemes, including the rational expectations one. We find that the usual argument in favor of the ‘rationality’ of rational expectations is a partial equilibrium one and that, taking into account the full general equilibrium consequences, rational expectations are dominated (in terms of individual utilities) by ‘non-rational’ expectations.