I attempt to present, explicate, and criticize Tarski’s semantic conception of truth. While it is a technical achievement of the first magnitude, it does not offer us an adequate account of truth. The most it can do is extensionally to offer us a frozen crosssection of language in a particular application. It cannot come to terms with the inflexibility and imprecision so important to the actual use and effectiveness of a natural language. Rather, as Davidson shows, given a pre-analytic conception of truth, semantic analyses can be more useful in exploring the meaning-structure of a language. An adequate account of truth must take stock of the pragmatics of person, purpose, and circumstances.