The question of whether there have been population size bottlenecks within the past million years was raised by the application of genetic data to human paleodemography, with the finding that human mtDNA has little variation relative to the current size of the human population (Cann, Stoneking, and Wilson 1987 ; Excoffier 1990 ; Vigilant et al. 1991 ). Many researchers used mtDNA diversity to estimate the time to the most recent common ancestor, or coalescence time, of human mtDNA. A wide range of estimates was obtained (fig. 5 ); 200,000 years is a widely accepted median estimate. This estimate allows an estimate of inbreeding Ne of about 8,800 individuals (table 1 ). The effective human population size estimated from mitochondrial diversity is therefore far removed from traditional estimates of the census population size of our species in the past (Weiss 1984 ).
The small effective size of mtDNA led to the hypothesis that the human lineage had undergone a recent bottleneck in population size. It was suggested that at the time of mtDNA coalescence, the entire human species was limited to one thousand to several thousand individuals (Cann, Stoneking, and Wilson 1987
; Vigilant et al. 1991
). To account for the mtDNA data, such a bottleneck would have to have been of sufficient duration to allow the fixation by drift of a single ancestral mtDNA variant. Presumably, this bottleneck was followed by expansion to the current population size. The postbottleneck population expansion could have resulted in a relatively increased number of low-frequency genetic variants. This would explain the departure of mtDNA from neutral mutation-drift equilibrium (Excoffier 1990
; Merriwether et al. 1991
; William, Ballard, and Kreitman 1995
; Nachman 1996
; Hey 1997
; Loewe and Scherer 1997
; Parsons, Muniec, and Sullivan 1997
; Wise, Sraml, and Easteal 1998
). Proceeding from this expectation, several researchers have examined the possibility of recent population expansions, such as those that would follow a bottleneck, using the distribution of pairwise genetic differences in human mtDNA (Harpending et al. 1993
; Sherry et al. 1994
; Rogers and Jorde 1995
). This distribution appears to be consistent with a massive Late Pleistocene population expansion.
The hypothesis of a recent population size bottleneck is also supported by some analyses of the human Y chromosome (Dorit, Akashi, and Gilbert 1995, 1996 ; Hammer 1995 ; Whitfield, Sulston, and Goodfellow 1995 ; Underhill et al. 1997 ; Hammer et al. 1998 ). For the parts of the Y chromosome with observed variation, coalescence time estimates vary from 37 to 516,000 years (Hammer 1995 ; Donnelly et al. 1996 ; Fu and Li 1996 ; Weiss and von Haeseler 1996 ; Hammer et al. 1998 ). The antiquity of Y-chromosomal variation is not significantly different from that of mtDNA (Hammer 1995 ). As in the case of human mtDNA, estimated Ne for the human Y chromosome is low and is consistent with a recent period of small population size. However, if this is the result of a recent bottleneck, such a bottleneck would have to have been of sufficient duration to cause the fixation of a single Y chromosome variant. As with mtDNA, this bottleneck would be expected to cause a departure from equilibrium in the Y chromosome data. This expectation is apparently met by the frequency spectrum of Y chromosome variants (Harpending et al. 1998 ).
The interpretation that the departure from neutral mutation-drift equilibrium reflects population size expansions assumes selective neutrality for these gene systems. However, several geneticists have suggested that selection may influence the distribution of mtDNA and Y chromosome variation in humans (Whitfield, Sulston, and Goodfellow 1995 ; Hey 1997 ; Templeton 1997 ; Wise, Sraml, and Easteal 1998 ). This has been a persistent interpretation from studies examining haploid and autosomal variation in the same individuals. Within nonrecombining systems such as mtDNA and parts of the Y chromosome, all the alleles are linked, so selection on any portion reduces variability in the entire genome (Spuhler 1989 ; Braverman et al. 1995 ; Templeton 1997 ; Nachman et al. 1998 ). Genetic systems with little or no recombination are consistently biased toward low levels of variation in Drosophila. Selection is the only reasonable explanation for the pattern of interlocus variance in Drosophila (Nurminsky et al. 1998 ; McAllister and Charlesworth 1999 ), where regions with low rates of recombination retain greater intraspecific diversity than those with higher rates of recombination (Begun and Aquadro 1991, 1992 ; Hudson 1994, 1995 ; Stephan et al. 1998 ). The same pattern of variation is found on the human X chromosome (Nachman et al. 1998 ) and may characterize other parts of the human genome.
The suggestion that selection has occurred many times in human evolution is not unexpected, and it is consistent with the pattern of great morphological change in humans during the past 2 Myr. Selection could take several forms. Hitchhiking (Kaplan, Hudson, and Langley 1989 ; Johnson 1999 ) would help explain the small Ne calculated for these nonrecombining systems because of their linkage. Background selection (Charlesworth, Morgan, and Charlesworth 1993 ) is an alternative explanation for reduced variation that is related to selective sweeps, since hitchhiking during a selective sweep could be followed by background selection (Nachman et al. 1998 ). An explanation for low Ne based on selection is more compatible with the lack of ancient genetic variation in these systems than a short-duration ( bottleneck of very small population size. The possibility that there has been selection in these nonrecombining systems (Hudson 1994, 1995 ; Stephan et al. 1998 ; Whitehead 1998 ) points to the necessity of considering autosomal diversity in humans for further evidence of whether the hypothesis of a severe recent bottleneck that some interpretations of haploid variation suggest can be refuted.