Human genome-wide association studies (GWAS) of longevity attempt to identify alleles at different frequencies in the extremely old, relative to a younger control sample. Here, we apply a GWAS approach to "synthetic" populations of Drosophila melanogaster derived from a small number of inbred founders. We used next-generation DNA sequencing to estimate allele and haplotype frequencies in the oldest surviving individuals of an age cohort and compared these frequencies with those of randomly sampled individuals from the same cohort. We used this case-control strategy in four independent cohorts and identified eight significantly differentiated regions of the genome potentially harboring genes with relevance for longevity. By modeling the effects of local haplotypes, we have more power to detect regions enriched for longevity genes than marker-based GWAS. Most significant regions occur near chromosome ends or centromeres where recombination is infrequent, consistent with these regions harboring unconditionally deleterious alleles impacting longevity. Genes in regions of normal recombination are enriched for those relevant to immune function and a gene family involved in oxidative stress response. Genetic differentiation between our experimental cohorts is comparable to that between human populations, suggesting in turn that our results may help explain heterogeneous signals in human association studies of extreme longevity when panels have diverse ancestry.