DNA methylation is known to be associated with cancer susceptibility. Such biochemical process, however, is also affected by other factors such as age, tissues, nutrition and other environmental variates. In other words, the methylation pattern can vary greatly between and within individuals. An appropriate study design for DNA methylation, therefore, should be able to control these sources of variation. Because most current case-control studies for identification of differentially methylated CpG sites may not be able to account for this heterogeneity, we propose in the present study for matched cases and controls a Bayesian hierarchical model with specially designed priors for CpG sites locating in different areas. This model can accommodate the individual heterogeneity in methylation data and allows the CpG sites to express non-exchangeable patterns. The analysis showed that this model can incorporate more biological interpretation with two different types of prior distributions considered for CpG islands and non-CpG enriched regions, respectively. The United Kingdom Ovarian Cancer Population Study (UKOPS) was used for illustration; methylation data from the study was generated by Illumina Infinium BeadArray technology. Parameters were estimated by Markov chain Monte Carlo (MCMC) method using OpenBUGS software package. The hyperparameter λ is of interest to measure methylation difference between case and age-matched control at each specific CpG. Probability of λ>0 in posterior samples was calculated for each CpG locus; 0.70, 0.90, 0.95, and 0.99 cut-off points of Pr (λ_i>0) resulted in 7877, 1068, 421, and 90 potential hypermethylated CpGs, respectively. A gene ontology analysis showed that 398 genes of hypermethylated CpGs in the 0.95 cutoff group were enriched in functions associated with carcinogenesis, including programmed cell death, positive regulation of cell cycle, and immune cell activation.