Taking advantage of increasingly available high-density single nucleotide polymorphism (SNP) markers within genes and across genomes, more and more genetic association studies began to use multiple closely linked markers in candidate genes. A practical analytical challenge arising in such studies is the possibility that not all case chromosomes have inherited disease-causing mutations from a common ancestral chromosome (founder heterogeneity). To alleviate the problem, we propose a method that applies a clustering algorithm to haplotype similarity analysis. The method identifies a sequence of nested subsets of case chromosomes by a peeling procedure, where each subset is relatively homogeneous. The average similarity score estimated from each subset in the sequence is compared to that estimated in controls, and a raw (unadjusted for multiple comparisons) P value is obtained. The test for the association between the trait and the candidate gene is based on the minimum raw P value observed in the comparison sequence, with its significance level estimated by a permutation procedure. The method can be applied to both haplotype and genotype data. Simulation studies suggest that our method has the correct type I error rate, and is generally more powerful than existing methods of haplotype similarity analysis.