The history of biology reveals numerous interpretations of the relationship between ontogeny and phylogeny, among them Garstang's critical realization that phylogeny can only be the history of changes in an ancestral ontogeny. One extrapolation of Garstang's conclusion is that the study of phylogeny should involve comparisons of ontogenies, in order to allow a reconstruction of the sequence of changes that led to the present variation in these ontogenies as well as the mechanisms underlying these changes. It is suggested that this can be done by recognizing homologous developmental stages among various ontogenies and subjecting these stages to a cladistic analysis. Some of the problems involving such analysis are explored in the course of examining the phylogeny of pharyngeal pouch ontogenies as a model system. It is concluded that sequences of ontogenetic stages are conserved (von Baerian recapitulation) but that terminal (Haeckelian recapitulation) and nonterminal alterations in the ancestral ontogeny are frequent. Both terminal and nonterminal alterations occur with approximately equal frequency in the case of pharyngeal pouches, but non-terminal deletion is rare to nonexistent. Examination of two additional morphological cases in which ontogeny has been utilized to infer phylogenetic changes-ontogenetic parcellation of neural systems and craniate head organization-reveals a number of problems from the point of view of a cladistical analysis of ontogenesis. Ontogenetic parcellation is essentially a recasting of Haeckel's biogenetic law and is rejected on the grounds that ontogenies do not parallel phylogenies, and only primitive and derived characters, not organisms, can be recognized. It is argued that the phylogenetic significance of transient neural characters can be determined only within a cladistical context and that insufficient ontogenetic data presently exist to evaluate most characters. Finally, the history of current models of craniate head organization is traced, and this information, along with that from more recent developmental studies, necessitates rejection of the Goodrich and Jarvik-Bjerring models. A new model is presented that purports to incorporate current ideas of somitomere and neuromere distribution, but it is deemed inadequate because, like previous models, it is based on information gleaned primarily from the ontogeny of a single species, whereas the model sought is a model of the ontogeny of ancestral craniates. To arrive at such a model, we must generate a morphotype, based on shared primitive features of cephalic ontogenies of all craniate radiations, which involves a cladistic analysis of these ontogenies.