The paper shows that, when dealing with social justice, it is necessary but not sufficient to consider the social evaluator as placed behind a veil of ignorance. Then, her judgements are conform to the moral principle of universalisation and thus are impartial. But the way such judgements should be modelled depends on the interpretation, whether utilitarian or Kantian, given to the universalisation principle. The article compares the utilitarian model of the impartial observer from Harsanyi with the Kantian model of the ignorant observer from Gajdos and Kandil. It defends the latter, which shows that, differently to what Rawls claimed, the maximin is not the unique rule of decision under ignorance. Classification JEL : B41, D60, D63