Joseph Raz’s account of norms provides that a norm requiring an agent to φ is a reason to φ protected by an exclusionary reason not to act on some other reasons. I present a dilemma concerning the determination of the contents of this set of excluded reasons. The question is whether or not the set includes reasons that count in favour of φing. If the answer is yes, the account is committed to a picture of norms that seems inconsistent with the phenomenology, in that it rules out acting on the very reasons on which the norm is based. If the answer is no, the account licenses a problematic form of double counting of reasons. I conclude that Raz’s protected reasons account of norms must be rejected, and tentatively suggest a solution to the problem posed by the dilemma.