Genomic imprinting refers to a pattern of gene expression in which a specific parent’s allele is either under-expressed or completely silenced. Imprinting is an evolutionary conundrum because it appears to incur the costs of diploidy (e.g. presenting a larger target than haploidy to mutations) while foregoing its benefits (protection from harmful recessive mutations). Here, we critically evaluate previously proposed evolutionary benefits of imprinting and suggest some additional ones. We discuss whether each benefit is capable of explaining both the origin and maintenance of imprinting, and examine how the different benefits interact. We then outline the many costs of imprinting. Simple models show that circulating deleterious recessives can prevent the initial spread of imprinting, even if imprinting would be evolutionarily stable if it could persist long enough to purge these. We also show that imprinting can raise or lower the mutation load, depending on the selective regime and the degree of dominance. We finish by discussing the population-level consequences of imprinting, which can be both positive and negative. Imprinting offers many insights into evolutionary conflict, the interaction between individual- and population-level fitness effects, and the ‘gene’s-eye view’ of evolution.