Infectious diseases now threaten wildlife populations worldwide but population recovery following local extinction has rarely been observed. In such a case, do resistant individuals recolonize from a central remnant population, or do they spread from small, perhaps overlooked, populations of resistant individuals? Introduced avian malaria (Plasmodium relictum) has devastated low-elevation populations of native birds in Hawaii, but at least one species (Hawaii amakihi, Hemignathus virens) that was greatly reduced at elevations below about 1000 m tolerates malaria and has initiated a remarkable and rapid recovery. We assessed mitochondrial and nuclear DNA markers from amakihi and two other Hawaiian honeycreepers, apapane (Himatione sanguinea) and iiwi (Vestiaria coccinea), at nine primary study sites from 2001 to 2003 to determine the source of re-establishing birds. In addition, we obtained sequences from tissue from amakihi museum study skins (1898 and 1948-49) to assess temporal changes in allele distributions. We found that amakihi in lowland areas are, and have historically been, differentiated from birds at high elevations and had unique alleles retained through time; that is, their genetic signature was not a subset of the genetic variation at higher elevations. We suggest that high disease pressure rapidly selected for resistance to malaria at low elevation, leaving small pockets of resistant birds, and this resistance spread outward from the scattered remnant populations. Low-elevation amakihi are currently isolated from higher elevations (> 1000 m) where disease emergence and transmission rates appear to vary seasonally and annually. In contrast to results from amakihi, no genetic differentiation between elevations was found in apapane and iiwi, indicating that slight variation in genetic or life-history attributes can determine disease resistance and population recovery. Determining the conditions that allow for the development of resistance to disease is essential to understanding how species evolve resistance across a landscape of varying disease pressures. ?? 2007 The Authors.