There is some sense of "ought" in which what an agent ought to do depends on her epistemic state--e.g., such that she ought to take whatever she justifiably regards as the best available course of action. Oughts of this kind are closely connected to action-guidance, since unlike "objective oughts" which are epistemic state-invariant, they seem to be epistemically accessible to agents under most circumstances. I argue, however, that under some conditions (namely, conditions of normative uncertainty) there may be no interesting species of oughts to which agents have epistemic access. This constitutes a challenge for any theory of rational choice that aims to provide agents with all-things-considered action guidance.