Respect for patient autonomy has become the preeminent principle of medical ethics, to the point that tools have been developed, such as instructive directives, in an attempt to preserve a semblance of autonomy even when it has become clearly and irretrievably lost. Much of the practice around the respect for autonomy, however, mistakenly supposes that the capacity for autonomous choice is an all-or-nothing proposition. But seriously ill patients may retain some ability to participate in discussions of medical care yet have their autonomy profoundly compromised by physical duress, cognitive dysfunction, or delirium. The choices of individuals with compromised autonomy do not carry the same moral weight as those of the fully autonomous. Clinicians, therefore, cannot rely on such choices for guiding medical decisions and are obligated to evaluate them more fully before acting. We argue that clinicians should compare the choices of individuals with compromised autonomy to a medical assessment of the patient’s best interest. When the patient’s choice and the best-interests assessment are discordant, acting in the patient’s best interest may, at times, rightly override the requests of the patient. Such an approach, under a tightly constrained set of circumstances, would permit both the provision and the withholding of medical interventions despite patient requests to the contrary.