During the last few years, the regionalization of the rostral-caudal axis has been extensively studied through treatments with RA and genetic manipulations of Hox-C genes. RA shifts several Hox expression boundaries rostrally, deletes anterior rhombomeres but expands the caudal ones, and induces homeotic transformations in the vertebral column. These phenotypes indicate that retinoids may act in a graded fashion in the A-P axis, with maximum activity caudally. This excludes forebrain and midbrain, which apparently depend on neither Hox-C genes nor RA modulations, at least during early development. The phenotypes resulting from ectopic overexpression and loss of function of Hox genes described so far show homeotic transformations in paraxial mesoderm derivatives but not in the neurectoderm. An explanation for this discrepancy implies that the paraxial mesoderm may be already segmented in molecular terms at the time of Hox activation. Conversely, the activation of a distinct Hox-code without a previous "molecular segmentation" may specify rhombomeres with their own boundaries. This would explain why RA expands but does not duplicate the postotic rhombomeres. Finally, the atavistic transformations obtained by overexpressing Hox genes in the wrong place suggest that evolution might be introducing modifications within Hox regulatory regions. Thus, changes in their expression domains could sustain phylogenetic requirements.