Comparison of polymorphic sites such as single nucleotide polymorphisms (SNPs) within a gene between cases and controls may be useful for establishing a role of this gene in disease susceptibility. The approach includes two steps: the first step is the discovery of the different SNPs within the candidate gene and the second step is the association testing per se that can be done on the entire set of sites discovered or on a subset of these sites only. Selecting a subset of sites may increase the power to detect the association with the candidate gene since a smaller number of tests would then be performed. We proposed a strategy to select sites within a candidate gene and applied it on the Genetic Analysis Workshop 12 candidate gene data. Using these selected sites, we detected an association with candidate genes 1 and 6.