Our perception of natural killer (NK) cells has transformed dramatically since their discovery just over 40 years ago.1–۵ Once viewed as a monolithic fraction of peripheral blood lymphocytes with an innate ability to find and kill aberrant cells, NK cells have since been revealed as a highly diverse lymphocyte population within which idiosyncratic education, regulation and activation processes govern subset function and prominence. Subsets distinguished by variable constellations of regulatory receptors diverge further along functional scales delineated by the presence and affinity of polymorphic ligands. Genetic, stochastic, deterministic and epistatic factors interact to produce up to 30 000 discrete NK cell subsets.6 We now know that the ontogenetic NK cell repertoire is dynamically modified by environmental influences (Fig. 1). As with the responses of B and T lymphocytes, NK cell engagement in an immune response reshapes the baseline repertoire, in part by introducing an adapted, preferentially selected, memory-like population with distinct phenotypic and functional features. Selective expansion, together with acquisition of novel functional and phenotypic characteristics, is common to memory B, T and NK cells alike. However, to our knowledge, only NK cells mature into immune memory cells independently of antigen recognition through somatically generated, clonotypic receptors. Therefore, the selection mechanisms underlying dynamic remodelling of NK cell repertoires and NK cell maturation into memory cells remain mostly obscure. This review will address accumulating evidence for antigen-dependent and -independent NK cell maturation and memory formation. The response against murine cytomegalovirus (MCMV) is the first and best characterized example of antigen-driven activation and maturation of NK cells into memory cells. We will review aspects of the NK cell response to MCMV, followed by a series of as yet mechanistically unexplained examples of NK cell memory and antigen specificity in mice and primates, followed by a description of human memory-like NK cell development, primarily in response to human CMV (HCMV) infection. Of note, the process by which NK cell memory forms in the case of HCMV infection appears to have less in common with MCMV infection than initially expected. Finally, we will speculate on how features of NK cells, viruses and other antigens might interact to confer a semblance of antigen specificity independent of somatically rearranged clonotypic receptors.