Jason Stanley has tried to revive the description theory of reference for natural kind terms by restricting it to experts. On his view, being an expert about elms requires having uniquely identifying knowledge about elms, knowledge that is conventionally associated with the term ‘elm’ and that fixes its reference. But Stanley's attempt fails. For, as I show, even experts about elms can lack such knowledge by being epistemically irresponsible, mistaken or just plain cautious. The description theory of reference is false even when restricted to the most expert users of a term.