The equipoise requirement in clinical research demands that, if patients are to be randomly assigned to one of two interventions in a clinical trial, there must be genuine doubt about which is better. This reflects the traditional view that physicians must never knowingly compromise the care of their patients, even for the sake of future patients. Equipoise has proven to be deeply problematic, especially in the Third World. Some recent critics have argued against equipoise on the grounds that clinical research is fundamentally distinct from clinical care, and thus should be governed by different norms. I argue against this "difference position," and instead take issue with the traditional, exclusively patient-centered account of physicians' obligations that equipoise presupposes. In place of this traditional view, I propose a Kantian test for the reasonable partiality that physicians should show their patients, focusing on its application in clinical research and medical education.