This Article critically examines the core infringement standard for trademark law—the “likelihood of confusion” test—which imposes liability if an appreciable number of consumers are likely to be confused by the defendant’s use of its mark. The test is a mess. It produces bad results, is doctrinally incoherent, and lacks a sensible normative foundation. It chills socially valuable uses and facilitates excessively broad expansions of trademark law, and it includes factors that make no sense as predictors of likely confusion. Most importantly, the test erroneously assumes that consumer confusion is enough to warrant liability without regard to the harmful consequences or the moral wrongfulness of the defendant’s conduct. This Article diagnoses these problems and proposes a set of promising reforms. It begins by presenting a systematic history of the test’s origins. The likelihood of confusion test emerged in the early 1960s as a compromise between conflicting views about the proper scope of protection against noncompeting uses, a compromise that papered over the conflict without resolving the underlying normative disagreement. The result is a test without a secure normative foundation and one that remains plagued by the conflicts and disagreements that gave it birth. The test’s history also suggests a promising avenue for reform. To improve the infringement standard, one must start at the normative level and work from a coherent theory of trademark law’s goals. This Article distinguishes between moral and economic goals and fits infringement standards to the most plausible justifications for protecting marks. In those cases involving a moral wrong, such as intentional deception, moral principles should shape the test. In those cases where the economic goals of trademark law are paramount, the test should focus not only on the probability of confusion but also on the trademark-related harm that confusion generates.