Aberrant DNA methylation has long been recognized to be associated with human diseases. In this talk, I will discuss statistical challenges in mediation analyses for dissecting the role of DNA methylation in epidemiologic studies. The first example is testing for mediation of an environmental exposure on a disease outcome by epigenetic markers such as CpG methylation. When there are epigenome-wide CpGs being evaluated for mediation, the classical max-P procedure for the composite null hypothesis of mediation analysis can be overly conservative. The second example is assessing mediation of DNA methylation in genetic regulation of gene expression, i.e. expression quantitative trait locus (eQTL). When the temporal order of a candidate mediator and an outcome cannot be established, current methods are generally inadequate to distinguish three possible tri-variate relationships: mediation, passive consequence and independent association. I will present novel statistical testing methods to properly control for type I errors in high-dimensional mediating analyses (the first example), and to effectively test for DNA methylation being mediation or passive consequence of eQTL regulation.