Humans exhibit characteristic success in considering what is relevant in their cognitive tasks. Yet understanding how relevance is determined in cognition remains a problem. This paper seeks to make headway on this problem. The relevance problem is first introduced. Sperber and Wilson's influential theory of relevance is then discussed, but dismissed as inadequate. Some conditions are identified that an adequate definition of relevance might reasonably be expected to satisfy. A novel way to conceive of relevance is suggested which proves to be useful in understanding human cognitive performance.