26When a species adapts to a new habitat, selection for the fitness traits often result in a 27 confounding between genome-wide genotype and adaptive alleles. It is a major statistical 28 challenge to detect such adaptive polymorphisms if the confounding is strong, or the effects 29 of the adaptive alleles are weak. Here, we describe a novel approach to dissect polygenic 30 traits in natural populations. First, candidate adaptive loci are identified by screening for loci 31 that are directly associated to the trait or control the expression of genes known to affect it. 32Then, the multi-locus genetic architecture is inferred using a backward elimination 33 association analysis across all the candidate loci using an adaptive false-discovery rate based 34 threshold. Effects of population stratification are controlled by corrections for population 35 structure in the pre-screening step and by simultaneously testing all candidate loci in the 36 multi-locus model. We illustrate the method by exploring the polygenic basis of an important 37 adaptive trait, flowering time in Arabidopsis thaliana, using public data from the 1,001 38 genomes project. Our method revealed associations between 33 (29)