There has been a common presumption in debates surrounding social construction that to catch out some entity or category as so constructed is at the same time to condemn it. Thus Ian Hacking notes that "a primary use of 'social construction' is for consciousness raising;" it is "critical of the status quo." Social constructionists generally move, Hacking argues, from the argument that a given entity or category X "need not have existed," to the view that "[w]e would be much better off if X were done away with, or at least radically transformed" (Hacking 1999, p. 6). On this line of thinking, every entity or category is expected either to be a real feature of the world, something left over when the world is carved at its joints, or it is to be exposed as constructed and by the same measure to be relegated to the scrap-heap along with phlogiston, the ether, and so on. But the category of 'race' seems to defy this dichotomy. Since the mid-20th century no mainstream scientist has considered race a biologically significant category; no scientist believes any longer that 'negroid', 'caucasoid' and so on represent real natural kinds, carve nature at its joints, and so on. For several decades it has been well established that there is as much genetic variation between two members of any supposed race, as between two members of supposedly distinct races (see in particular Lewontin 1972). This is not to say that there are no real differences, some of which are externally observable, between different human populations; it is only to say, in Lawrence Hirschfeld's words, that "races as socially defined do not (even loosely) capture interesting clusters of these differences" (Hirschfeld 1998, p. 4). And yet the category of race continues to be deployed in a vast number of contexts, and certainly not just by racists, but by ardent antiracists as well, and by everyone in between. The history of race, then, is not like the history of phlogiston: an entity that is shown not to exist and that accordingly proceeds to go away. How are we to explain this difference? This is the principal question I would like to consider in the present article.