During the past half-century, a set of statistical techniques and ideas about inference have experienced a remarkable scientific success. Significance at the 5 percent level has come to mark a clear and distinct criterion for scientific knowledge in a wide range of fields. Recently, however, this convention has been embroiled in controversy, as the relentless pursuit of significance has produced a range of well-known scientific abuses. Instead of staking out a position in these debates, this article analyzes the history of epistemological values underlying them. It focuses on an earlier critic of the misuse of statistical tests: John W. Tukey. Speaking to behavioral scientists in the middle of the twentieth century, Tukey insisted that reducing inference to a set of universal rules or mechanical procedures to eliminate uncertainty was a pursuit doomed to failure. Scientists needed to accept the irreducibility of individual judgments and decisions in data analysis, even when they risked charges of subjectivism or arbitrariness. For Tukey, the enforcement of scientific consensus and even the value of objectivity must yield to empirical judgments and an ethic of individual conscience. These values were informed by his comparative understanding of the history of science, which reserved a special place for empiricism in younger sciences. Reconstructing Tukey’s work offers an alternative perspective on the quantitative, formal objectivity of the postwar sciences as well as the present, where big data and machine learning have raised thorny new problems for statistical inference and scientific expertise.