Epigenetic variation contributes to explaining the missing heritability of complex traits. In order to understand the genome-wide methylation variation in spring barley, our objectives were to gain fundamental insight into the barley methylome through whole genome bisulfite sequencing, characterizing methylation variation among 23 parental inbreds of a community resource for genetic mapping of phenotypic traits, and assessing the association of differentially methylated regions (DMRs) with single nucleotide polymorphisms (SNPs) and gene expression variation. Compared to other angiosperms, barley was found to have a highly methylated genome with an average genome wide methylation level of 88.6%, 58.1%, and 1.4% in the CpG, CHG, and CHH sequence context, respectively. We identified just below 500000 differentially methylated regions (DMRs) among the inbreds. About 64%, 64%, and 83% of the DMRs were not associated with genomic variation in the CpG, CHG, and CHH context, respectively. The methylation level of around 6% of all DMRs was significantly associated with gene expression, where the directionality of the correlation was depended on the relative location of the DMR to the respective gene with a recognizable pattern. Notably, this pattern was much more specific and spatially confined than the association of methylation with gene expression across genes in a singular inbred line. We exemplified this association between DNA methylation and gene expression on the known flowering promoting gene VRN-H1 and identified a highly methylated epiallele associated with earlier flowering time. Finally, methylation was shown to improve the prediction abilities of genomic prediction models for a variety of traits over models using solely SNPs and gene expression as predictors. These observations highlight the independence of DNA methylation to sequence variation and their difference in information content. Our discoveries suggest that epigenetic variation provides a layer of information likely not predictable by other means and is therefore a valuable addition to genomic prediction models.