Pinpointing culprit causal variants along signal peaks of genomewide association studies (GWAS) is challenging. To overcome confounding effects of multiple independent variants at such a locus and narrow the interval for causal allele capture, we developed an approach that maps local shared haplotypes harboring a putative causal variant. We demonstrate our method in an extreme isolate founder population, the pacific Island of Kosrae. We analyzed plasma plant sterol (PPS) levels, a surrogate measure of cholesterol absorption from the intestine, where previous studies have implicated 2p21 mutations in the ATP binding cassette subfamily G members 5 or 8 (ABCG5 or ABCG8) genes. We have previously reported that 11.1% of the islanders are carriers of a frameshift ABCG8 mutation increasing PPS levels in carriers by 50%. GWAS adjusted for this mutation revealed genomewide significant signals along 11 Mb around it. To fine-map this signal, we detected pairwise identity-by-descent haplotypes using our tool GERMLINE and implemented a clustering algorithm to identify haplotypes shared across multiple samples with their unique shared boundaries. A single 526-kb haplotype mapped strongly to PPS levels, dramatically refining the mapped interval. This haplotype spans the ABCG5/ABCG8 genes, is carried by 1.8% of the islanders, and results in a striking 100% increase of PPS in carriers. Resequencing of ABCG5 in these carriers found a D450H missense mutation along the associated haplotype. These findings exemplify the power of haplotype analysis for mapping mutations in isolated populations and specifically for dissecting effects of multiple variants of the same locus.genetics ͉ genomewide association study ͉ haplotype mapping ͉ plasma plant sterols R apid technological advances in genomics over the last few years have focused on identifying common genetic variants affecting complex disease risk. To date genome-wide association studies (GWAS) have reproducibly implicated over 270 genomic regions that modify the risk for over 70 complex disease and health-related phenotypes (1-3). However, identifying the culprit causal variant underlying these local peaks in association signals remains a challenging task with only a handful of causal variants identified so far (4 -6). One conventional approach is sequencing a region of arbitrary length around the signal peak across enough individuals to capture the causal allele. However, the larger the sequenced interval and number of individuals, the more variants discovered making it difficult to distinguish the driver allele from the large number of passenger variants. This problem is exacerbated in cases where multiple, seemingly independent signals reside in close proximity along the genome. Recent examples of multiple alleles include the 8q24 region for prostate cancer (7-9), 6p21 region for HIV-1 viral setpoint (10), the IL23R, 6p25, and 17q21 regions for Crohn's disease (11, 12), the PNPLA3 region and nonalcoholic fatty liver disease (13), the IRF5, STAT4, and TNFAIP3 regions for systemic...