Philosophers have traditionally inclined toward one of two opposite extremes when it comes to metaphor. On the one hand, partisans of metaphor have tended to believe that metaphors do something different in kind from literal utterances; it is a ‘‘heresy,’’ they think, either to deny that what metaphors do is genuinely cognitive, or to assume that it can be translated into literal terms. On the other hand, analytic philosophers have typically denied just this: they tend to assume that if metaphors express any genuine content at all, then that content can in principle be paraphrased into literal terms. They often conclude on this basis that metaphor is theoretically dispensable, and so that it poses no special challenges and affords no distinctive insights for the philosophy of language and mind. In this paper, I want to steer between these two extremes. Metaphors don’t do anything different in kind from what can be done with literal speech. But this does not render metaphor theoretically dispensable or irrelevant. In certain circumstances, I will argue, metaphors can enable speakers to communicate contents that cannot be stated in fully literal and explicit terms. These cases thus serve as counterexamples to the ‘‘Principle of Expressibility,’’ the idea that whatever can be meant can be said. Indeed, I will argue, the point goes for cognition as well as communication: metaphors can sometimes provide us with our only cognitive access to certain properties. In the end, I think, thinking about metaphor is useful because it draws our attention to patterns and processes of thought that play a pervasive role in our ordinary thought and talk, and that can genuinely extend our basic communicative and cognitive resources.