It is often argued that belief is partly constituted by a norm of truth. Most recent discussions have assumed that the norm is deontic concerning what may or ought to be believed. I criticize two proposals, one canvassed by Krister Bykvist and Anandi Hattiangadi, and the other defended by Daniel Whiting. Instead, I argue in favour of an evaluative norm, according to which we would do well to believe the truth. I show that an evaluative norm fares better than its deontic competitors with respect to the demandingness of truth, the aim of truth, and epistemic blame.