BACKGROUND : Thousands of mammalian promoters are defined by co-enrichment of the histone tail modifications H3K27me3 (repressive) and H3K4me3 (activating) and are thus termed bivalent. It was previously observed that bivalent genes in human ES cells (hESC) are frequent targets for hypermethylation in human cancers, and depletion of DNA methylation in mouse embryonic stem cells has a marked impact on H3K27me3 distribution at bivalent promoters. However, only a fraction of bivalent genes in stem cells are targets of hypermethylation in cancer, and it is currently unclear whether all bivalent promoters are equally sensitive to DNA hypomethylation and whether H3K4me3 levels play a role in the interplay between DNA methylation and H3K27me3. RESULTS : We report the sub-classification of bivalent promoters into two groups-promoters with a high H3K27me3:H3K4me3 (hiBiv) ratio or promoters with a low H3K27me3:H3K4me3 ratio (loBiv). HiBiv are enriched in canonical Polycomb components, show a higher degree of local intrachromosomal contacts and are highly sensitive to DNA hypomethylation in terms of H3K27me3 depletion from broad Polycomb domains. In contrast, loBiv promoters are enriched in non-canonical Polycomb components, show lower intrachromosomal contacts and are less sensitive to DNA hypomethylation at the same genomic resolution. Multiple systems reveal that hiBiv promoters are more depleted of Polycomb complexes than loBiv promoters following a reduction in DNA methylation, and we demonstrate that H3K27me3 re-accumulates at promoters when DNA methylation is restored. In human cancer, we show that hiBiv promoters lose H3K27me3 and are more susceptible to DNA hypermethylation than loBiv promoters. CONCLUSIONS : We conclude that bivalency as a general term to describe mammalian promoters is an over-simplification and our sub-classification has revealed novel insights into the interplay between the largely antagonistic presence of DNA methylation and Polycomb systems at bivalent promoters. This approach redefines molecular pathologies underlying disease in which global DNA methylation is aberrant or where Polycomb mutations are present.