This article incorporates self-fulfilling mistakes into an otherwise classical decision-making framework. A behavioural agent makes a mistake when he fails to internalise all the consequences of his actions on himself. We show that Sen's axioms α and γ fully characterise choice data consistent with behavioural agents. These two axioms are weaker than Sen's axioms α and β that fully characterise rational agents. We offer a welfare benchmark that can be applied to existing behavioural economics models and show the conditions under which our welfare ranking can be used to infer welfare dominated (i.e. mistaken) choices using choice data alone.