Abstract: Many people vigorously defend particular institutional judgments on such issues as the filibuster, recess appointments, executive privilege, federalism, and the role of the courts. These judgments are defended publicly with great intensity and conviction, but some of them turn out to be exceedingly fragile, in the sense that their advocates are prepared to change their positions as soon as their ideological commitments cut in the other direction. For example, institutional flip-flops can be found when Democratic officials, fiercely protective of the filibuster when the President is a Republican, end up rejecting the filibuster when the President is a Democrat. Other flip-flops seem to occur when Supreme Court justices, generally insistent on the need for deference to the political process, show no such deference in particular contexts. Our primary explanation is that many institutional flip-flops are a product of “merits bias,” a form of motivated reasoning through which short-term political commitments make complex and controversial institutional judgments seem self-evident (thus rendering those judgments vulnerable when short-term political commitments cut the other way). We offer evidence to support the claim that merits bias plays a significant role. At the same time, many institutional judgments are essentially opportunistic and rhetorical, and others are a product of the need for compromise within multimember groups (including courts). Judges might join opinions with which they do not entirely agree, and the consequence can be a degree of institutional flip-flopping. Importantly, some apparent flip-flops are a result of learning, as, for example, when a period of experience with a powerful president, or a powerful Supreme Court, leads people to favor constraints. In principle, institutional flip-flops should be reduced or prevented through the adoption of some kind of veil of ignorance. But in the relevant contexts, the idea of a veil runs into severe normative, conceptual, and empirical problems, in part because the veil might deprive agents of indispensable information about the likely effects of institutional arrangements. We explore how these problems might be overcome.