Retinoids, the active metabolites of vitamin A, regulate complex gene networks involved in vertebrate morphogenesis, growth, cellular differentiation and homeostasis. Studies performed in vitro, using either acellular systems or transfected cells, have shown that retinoid actions are mediated through heterodimers between the RAR and RXR nuclear receptors. However, in vitro studies indicate what is possible, but not necessarily what is actually occurring in vivo, because they are performed under non-physiological conditions. Therefore, genetic approaches in the animal have been be used to determine the physiological functions of retinoid receptors. Homologous recombination in embryonic stem cells has been used to generate germline null mutations of the RAR- and RXR-coding genes in the mouse. As reviewed here, the generation of such germline mutations, combined with pharmacological approaches to block the RA signalling pathway, has provided genetic evidence that RAR/RXR heterodimers are indeed the functional units transducing the RA signal during prenatal development. However, due to (i) the complexity in "hormonal" signalling through transduction by the multiple RARs and RXRs, (ii) the functional redundancies (possibly artefactually generated by the mutations) within receptor isotypes belonging to a given family, and (iii) in utero or early postnatal lethality of certain germline null mutations, these genetic studies have failed to reveal all the physiological functions of RARs and RXRs, notably in adults. Spatio-temporally-controlled somatic mutations generated in given cell types/tissues and at chosen times during postnatal life, will be required to reveal all the functions of RAR and RXR throughout the lifetime of the mouse.