ABSTRACTA fundamental component of molecular evolution are the rules that govern when, and why, a given change (allele) is deleterious or neutral. The ability to define such rules for epialleles – analogous to the rules at the DNA sequence level – would thus have profound implications for our understanding of epigenetic variation and evolution. Here, we focus on promoter methylation in the male human germline, which – apart from its role in gene regulation – is also known to greatly increase the mutation rate of CpG dinucleotides. We first develop a simple but general approach for detecting selection on epialleles, which does not require populationscale data. We then show that germline promoter methylation is deleterious at loss-of-function intolerant genes, but neutral at loss-of-function tolerant ones. In concordance with this, a human-mouse comparative analysis of sperm methylomes reveals strong suppression of methylation acquisition at loss-of-function intolerant promoters. We demonstrate that this selection is neither a secondary consequence of germline gene expression levels, nor of promoter H3K4me3 levels. Rather, the deleteriousness of promoter methylation is explained by its mutagenic effect on the underlying CpGs. Our results thus address a long-standing open question in molecular evolution, providing the first demonstration of selection acting on an epigenetic mutation rate modifier to locally dictate the mutation rate in humans. They also suggest the existence of a mechanism that preferentially protects loss-of-function intolerant promoters from methylation in the germline. Finally, they directly refute the prevailing dogma that CpG islands are not under active selection.