In his 1998 book The g Factor, scholar Arthur Jensen used an extremely objective method to classify humans in different races: Varimax rotation of principal components applied to the genes of 42 populations studied by scholars Nei & Roychoudhury (1993).  What I love about this is Jensen used the eigenvalues > 1 rule for determining the number of components to be retained for rotation.

So based on “genetic similarity”, there are roughly six major races, though in the chart below (which I adapted from Jensen) I prefer to label all six using just three major anthropological types: Negroid, Caucasoid, and Mongoloid.  Different ethnic groups have strong or weak loadings on the different components (races) and some load on multiple components, which as Jensen noted, reflect, central tendencies, not discrete categories.

varimax rotation components 1 2 3 4 5 6
population asian mongoloid caucasoid hybridized mongoloid african negroid american mongoloid non-african negroid
pygmy 651
nigerian 734
bantu 747
san (bushman) 465
lapp 500
finn 988
german 978
english 948
italian 989
iranian 635
northern indian 704
japanese 936 214
korean 959 229
tibetan 855
mongolian 842 357
southern chinese 331 771
thai 814
filipino 782
indonesian 749
polynesian 526 284
micronesian 521 328
australian (aborigines) 706
papuan (new guineans) 742
north amerindian 804
south amerindian 563
eskimo 726

I find it interesting that Italians are the most Caucasoid of all Caucasoid ethnicities (clocking in at 989). Italians look like a hybrid of what I believe are the three Caucasoid sub-races: whites, dark caucasoids, and Ashkenazim; thus it makes sense that Italians are the essence of the Caucasoid race.

It should be noted that genetic distance is measured using relatively neutral genes, which by definition are relatively insensitive to natural selection. It’s probably for this reason that sub-Saharan and Australoids get divided into different races, despite both being Negroid in appearance and IQ; because neutral genes primarily reflect the genetic clock (time since populations split) and not how truly similar folks are. So if you define race by how recently people shared a common ancestor, then this analysis should please you, but if you define race by how much people preserve the traits of a common ancestor, then an analysis using non-neutral genes is badly needed.