The US health care system is poised to learn more about preventing, diagnosing, and treating illness than has ever been possible. This change is powered by the increasing commitment to comparative effectiveness research, increases in practice-based research, and the increasing availability of data arising from electronic health information systems to help patients, clinicians, and others understand who benefits from which treatments. Much can be learned by observing the outcomes of the varied decisions that clinicians and hospitals make. However, for many health care questions, it is important to intervene by systematically varying care, for instance by randomly selecting the order in which a new practice is introduced into different parts of a system or by randomly assigning different commonly used treatments to patients who are good candidates for all of the approaches. Indeed, random assignment would be important to ascribe causality to the change.