It has been proposed that technical systems that estimate patient choices and preferences should guide treatment when patients are incapacitated. However, the ethical debate about such Patient Preference Predictors needs conceptual clarity. We distinguish between choices and preferences, between types of preferences, and between preferences over treatments and preferences over how treatment decisions are made. These distinctions show that respect for autonomy can take several forms: respect for a person’s explicit choices, for her overall preferences, or for her most central and enduring values. We argue that while Predictors may enhance beneficence, and reduce uncertainty and conflict, they raise normative questions about which preferences should be estimated and respected. In particular, we question the moral relevance of hypothetical choices, as well as the legitimacy of idealizing preferences beyond restored competence. Preferences over decision methods can be instrumental to treatment outcomes but can also be intrinsic. The latter may sometimes conflict with preferences regarding outcomes.