This paper examines the implication of the habit forming hypothesis for welfare theory. When preferences are no longer given exogenously, but rather change in relation to past consumption and other economic variables it is crucial to understand whether they still represent an adequate expression of individual welfare. On this point the distinction between individuals who are capable to predict future taste (rational habit forming models) and those who are not (myopic habit forming models) assumes essential importance. First the paper analyses the main criteria proposed by myopic habit forming models for measuring individual and social welfare. The same sort of problem undermines them all: the need for information that cannot be inferred from individuals' behaviour. Consequently it is necessary to question the legitimacy of paternalistic intervention. These issues become meaningless if the rational habit forming hypothesis is accepted. Thanks to the ability of individuals to predict their future preferences, the results achieved by traditional welfare theory are preserved. The utility function, however, must respect some restrictive conditions (Strotz 1956, Pollak 1968). If these are not respected, then the rational habit forming hypothesis can generate exchanges that do no benefit both parties (Yaari 1977). Moreover, the rational habit forming process of preference formation will aggravate some inefficiencies of the economic system (Hanhel-Albert 1990).