Section 107 of the Copyright Act of 1976 establishes the affirmative defense to copyright infringement of “fair use,” by far the most enigmatic doctrine in U.S. copyright law and by far the most important. Without it, much of our economic and communicative action would constitute copyright infringement. Yet despite the importance of the fair use defense, and despite the enormous amount of scholarly attention that it has received, we continue to lack any systematic, comprehensive account of our fair use case law and the actual state of our fair use doctrine. Instead, our conventional wisdom derives from a small set of conventionally agreed-upon leading cases. This Article presents the results of the first empirical study of our fair use case law to show that much of our conventional wisdom about that case law is mistaken. Working from a data set consisting of all reported federal opinions that made substantial use of the section 107 four-factor test for fair use through 2005, the Article shows which factors and subfactors actually drive the outcome of the fair use test in practice, how the fair use factors interact, how courts inflect certain individual factors, and the extent to which judges stampede the factor outcomes to conform to the overall test outcome. It also presents empirical evidence of the extent to which lower courts either deliberately ignored or were ignorant of fair use doctrine set forth in the leading cases, particularly those from the Supreme Court. Based on these descriptive findings, the Article prescribes a set of doctrinal practices that will improve courts’ adjudication of the fair use defense.