Genome-wide association studies (GWASs) aim to detect genetic risk factors for complex human diseases by identifying disease-associated single-nucleotide polymorphisms (SNPs). The most commonly-used GWAS method is the SNP-wise-test approach, in which an association test is performed for each SNP, and then the p-values are adjusted for multiple testing. However, this approach is often lack of power after multiple testing adjustments due to a huge number (> 1 million) of tests in GWAS. To address this problem, we propose a model-based clustering via a mixture of Bayesian hierarchical models, which could borrow information across SNPs to group SNPs to different clusters having different mean genotype levels between cases and controls. Simulation studies and real data studies showed that the proposed model-based clustering outperformed SNP-wised-test approach.