The analysis of population genomic data of entire species can inform us about their evolution and conservation status. We leveraged detailed phenotypic and genomic data of nearly all living kākāpō, an endangered and culturally significant parrot endemic to Aotearoa New Zealand, to describe the genomic architecture of its feather color polymorphism. Green and olive feather coloration are at similar frequencies in the population, which currently numbers less than 250 birds. We further dissected the color phenotype, demonstrating that the two colors differ in their UV reflection due to feather structure. We use quantitative genomics methods to identify two genetic variants whose epistatic interaction can fully explain the species’ color phenotype. The associated genomic region and our phenotypic analyses suggest a potential link to the candidate geneLHX8, involved in tissue patterning, tissue differentiation, and bone morphogenesis. Our genomic forward simulations show that balancing selection constitutes a likely evolutionary mechanism for the polymorphism to become established in an ancestrally large effective population size, and to be maintained through a continuous population decline, including a severe bottleneck of the kākāpō population. The combination of our genomic, phenotypic, and simulation analyses lets us propose the role of extinct apex predators as the likely agent of such selection, making the color polymorphism in the kākāpō a “ghost of selection past”.Significance statementWe use genetic data and trait measurements to understand why kākāpō, a critically endangered Aotearoa New Zealand bird, can be either green or olive. We demonstrate that this color difference is also reflected in a difference in UV reflection due to feather structure, and that color can be predicted by two genetic variants on chromosome 8. We show via genetic simulation that it is likely that the rarest color is favored, leading to fluctuating frequencies of green and olive over time. We propose that past visual predators of kākāpō could most readily identify the most common color. Our study shows the power of genomic analyses to uncover past evolutionary events, and to infer their potential relevance for future conservation management.