The argument that we should believe in the entities our best theories postulate rests on the claim that their success would otherwise be miraculous, and it is weaker than its confident statement suggests. The history of science supplies a list of theories that were successful, in the sense of yielding accurate predictions over long periods, and whose central entities are no longer thought to exist; the luminiferous ether is the standard case and it is not alone. Realists have responded by restricting the commitment to those parts of a theory doing the predictive work, which is a reasonable move and one that is hard to apply prospectively, since identifying the load-bearing parts of a current theory is exactly what a later theory does for us. The position is defensible and considerably less triumphant than it was.