This paper discusses whether it can be known a priori that a particular term, such as ‘water’, is a natural kind term, and how this problem relates to Putnam’s claim that natural kind terms require an externalist semantics. Two conceptions of natural kind terms are contrasted: The first holds that whether ‘water’ is a natural kind term depends on its a priori knowable semantic features. The second – currently gaining in popularity among externalists – holds that ‘water’ is a natural kind term only if water in fact turns out to be a natural kind. In the paper I argue that the first conception of natural kind terms should be rejected, since it does not adequately account for instances in which a purported natural kind term fails to name a natural kind (familiar examples are ‘air’, ‘sand’ and ‘jade’). However, I argue further Putnam’s externalism depends on this first, traditional conception of natural kind terms. The externalist is left with a Hobson’s choice: Either hold on to the traditional conception of natural kind terms, with its inherent problems, or reject extemalism. I argue that we should in fact reject externalism and reconsider the possibility of giving a descriptivist account of the meaning of natural kind terms.