Genetic clustering studies have shown that despite the small proportion of genetic variation separating continental populations, it is possible to assign some (geographically separated and not recently admixed) individuals to their (or their ancestors') continents of origin, based on genetic data alone. Is ‘race' vindicated? In this talk I argue against such a conclusion. I begin by discussing three problems for a racial reading of clustering studies: (1) the grain of resolution problem, (2) the non-concordance between clustering studies, and (3) the clinal (gradual) distribution of genetic structure and diversity. Then I consider some arguments for racial naturalism, and I find a surprising amount of agreement between myself and my ‘opponents'. It turns out that much of the disagreement in the race debate turns on the definition of race adopted by its participants. As a consequence the best way to settle the race debate may be to settle the semantics. I suggest that this should be done in a way that is both historically sensitive and consonant with how race is understood outside of biology. The difficulty of such an endeavour has caused many to argue that race has no definition. I propose that we can indeed define race, but only if we separate that definition from the question of how race is constructed. On my proposed – and, I hope, well-grounded – definition of race, genetic clustering studies do not support racial naturalism. I propose that our racial categories are best understood through an approach I call ‘interactive constructionism'.