Risk adjustment is intended to minimize selection of patients or enrollees in health plans. Current efforts generally are recognized as inadequate, but improvement is difficult. The greatest short-term gain will come from introducing diagnostic information, though outpatient diagnosis data are unreliable. Initial efforts may use inpatient data, but this creates incentives to hospitalize people. Even exploiting diagnosis information leaves substantial imperfections. Partial capitation, common in behavioral health, reduces incentives to select patients and stent on services, but current policy resists it, perhaps because policymakers misinterpret the lesson of the Prospective Payment System. Theoretically, not paying plans more for providing additional services is optimal only if consumers are well informed.