Realists about modality offer an attractive semantics for modal discourse in terms of possible worlds, but standard accounts of the worlds—as properties, propositions, or causally-isolated concreta—invoke entities with which we can’t interact. If realism is true, how can we know anything about modal matters? Let's call this "the Benacerraf Problem." I suggest that C. I. Lewis has an intriguing answer to it. Given that we’re willing to disentangle some of Lewis’s insights from his phenomenalism, we can take the following line. (...) If the Benacerraf Problem is a genuine one, then it threatens all knowledge—not just modal knowledge. But then it leads to a general and implausible form of skepticism, not a limited and more plausible debunking argument. Hence, whatever we’re willing to say about skepticism we should say about the Benacerraf Problem. (shrink)