Imagine a coin that lands heads 90% of the time. Flip it ten times: you'll likely see 9 heads, 1 tail. Flip it ten million times: you'll see almost exactly 9,000,000 heads, 1,000,000 tails. The deterministic signal of the biased coin emerges only at large sample sizes; in small samples, random fluctuation dominates. Evolution works the same way: natural selection is the biased coin, but in small populations over short timescales, random sampling — which individuals happen to reproduce, which gametes happen to fuse — can dominate selection, with allele frequencies drifting randomly to fix (100%) or be lost (0%) for reasons that have nothing to do with fitness. Genetic drift — the term coined by Sewall Wright in 1931 — is the non-selective evolutionary force that operates alongside selection, mutation, and migration.
Genetic drift refers to changes in allele frequency over generations due to random sampling, and its strength is inversely proportional to population size: in a population of N individuals (more precisely, effective population size Nₑ), the expected change in allele frequency per generation is on the order of 1/√Nₑ, so a population of 100 fluctuates noticeably each generation while a population of 100 million barely budges. Two important corollaries: every allele eventually fixes or is lost in a finite population if no other forces act (drift is a one-way ratchet toward homozygosity), and the probability that a new neutral allele eventually fixes equals its initial frequency, with expected fixation time of approximately 4Nₑ generations. The Kimura neutral theory (1968) is the deep claim built on this — most genetic variation within and between species is selectively neutral, with the molecular clock reflecting drift rather than selection — and has been used to date evolutionary events from species divergences to pathogen outbreaks. Founder effects show drift in concentrated form: when a small group establishes a new population, the new allele frequencies reflect the founders, and the Amish, Ashkenazi Jews, Finns, Afrikaners, and many island populations show elevated frequencies of specific genetic conditions (Tay-Sachs in Ashkenazi populations, Huntington's in some Venezuelan villages, BRCA1 mutations in Icelanders). Bottlenecks operate similarly: the cheetah (~7,000 individuals worldwide today, extremely low genetic diversity) appears to have undergone a severe bottleneck about ~10,000 years ago, and humans show a clear bottleneck signature about ~70,000 years ago (effective population size of a few thousand, consistent with the Toba supereruption hypothesis). The relative strength of drift versus selection in any case depends on population size, allele effect size, and timescale — strong selection on a large-effect allele in a large population fixes the allele predictably, while weak selection on a small-effect allele in a small population may have outcomes determined entirely by drift.
Population-genetic drift is now central to medical genetics: the Genome Aggregation Database (gnomAD) catalogues allele frequencies across human populations, and the substantial differences between populations are mostly attributable to drift since the migration out of Africa ~70,000 years ago plus founder effects, with the clinical implication that polygenic risk scores derived from European populations underperform in non-European populations. Conservation genomics uses drift-aware models to design captive breeding and reintroduction plans for endangered species (Florida panthers, California condors, black-footed ferrets, Tasmanian devils). SARS-CoV-2 variants during the pandemic were a real-time demonstration of drift + selection. Ancient DNA studies (David Reich lab) reconstruct human population movements over the past 50,000 years using drift signatures, with the resulting picture one of much more migration, mixing, and replacement than archaeology alone had suggested. The Wright-Fisher model that formalized drift in the 1930s remains the foundation of population genetics.