Results of association studies using individual single nucleotide polymorphisms (SNPs) or SNPhaplotypes have been inconsistent. Possible reasons could be attributed to poor experimental design, generalization of results from a single population or inappropriate choice of markers. Here we explore the possibility that the sequence context of a SNP may be responsible for its poor association with the phenotype. An analysis of the Human_MitBASE data helped in the prediction of association between SNP haplotypes with disease phenotypes. A novel computational tool E-MIDAS was developed to automate this analysis. Based on our results, we propose omission of SNPs in CpG dinucleotides which have a mutation predisposing flank and those present at sites of recurrent mutation, from association studies.