One desired goal for future retinoid development is the dissociation of therapeutic benefit from teratogenic risk. It is not known if this result can be achieved, but increasing our understanding of the molecular mechanisms of retinoid action offers rational approaches to the attainment of this goal. The basis for these new approaches is the recent discovery of nuclear receptors for all-trans-retinoic acid and for the 9-cis form of retinoic acid. These receptors, like the steroid hormone receptors, are ligand-dependent transcription factors. Retinoids act by binding to these receptors and modulating specific gene pathways that ultimately control cell differentiation and development. Retinoic acid is required for normal embryonic development, and retinoic acid concentrations are regulated in the developing embryo. The malformations associated with in utero exposure to retinoids might be the result of inappropriate activation of specific receptors and inappropriate modulation of specific gene pathways. At least three nuclear receptors for all-trans-retinoic acid and three nuclear receptors for the 9-cis form of retinoic acid are known to exist, with the two receptor subfamilies interacting to form heterodimers. This multiplicity of receptors and receptor interactions suggests that differences may exist among receptors in the biologic effects they mediate. Although some retinoid ligands are not specific, others are highly selective for one particular receptor. Contrasting the effects of ligands with different specificities should increase understanding of the biologic responses associated with specific receptors and efforts to dissociate therapeutic activity from undesired effects.