Economists test to strengthen conviction, though mostly without disclosing how this can occur. There is a clear line of logical implication from theory to model; but in the process there may be a narrowing and specialization of the hypothesis so that it is not clear what weight should be placed upon the result of a test on the model. Empirical models and tests may be narrowed down so far that they involve certain nonunique characteristics of the original theories. Even when a hypothesis is confirmed, no reverse inference back to theory can be made from such tests. We illustrate how a number of such 'characteristics tests' worked, drawing on American economists' tests of the natural rate hypothesis in the 1970s. © 1995.