This paper evaluates the recent trend to renounce the similarity approach to counterfactuals in favor of the older metalinguistic theory. I try to show, first, that the metalinguistic theory cannot work in anything like its present form (the form described by many in the last decade who claim to be able to solve Goodman's old problem of cotenability). This is so, I argue, because the metalinguistic theory requires laws of nature of a sort that we (apparently) do not have: current physical theory cannot underwrite the metalinguistic theory. Second, I draw from the first point a motivation for the similarity approach, a motivation based on theoretical considerations apart from the standard ones of pretheoretical intuition.