Any phylogenetic investigation starts with a mental decomposition of the organisms into units of description or characters. Only then can the techniques to evaluate the historical relationships among character states be applied and genealogical continuity inferred. Character definition is expected to be non-arbitrary, such that the union of a hoof and the cerebellum is not acceptable as a character. Instead describing the shape of a claw, or the location of the nasal opening are acceptable units of description. It is implicitly expected that characters and then homologues are natural units, but no agreement has been reached about of what kind these natural units shall be. This is the main reason why the homology concept is so elusive (Wagner, 1995). To overcome this elusiveness it is necessary to find out how natural kinds or natural units are recognized.
An interesting answer to this question was provided by Willard V. Quine in his seminal essay on Natural Kinds (1969). Paradigms of natural kinds are atoms, genes and species. Quine compared various approaches to define natural kinds, using similarity or statistical approaches but concluded that neither of them is suitable. He finally suggested that natural kinds can only be defined in the context of a process or a theory of a process in which these entities act as a unit. For instance atoms act as units in chemical reactions, genes are the units of genetic transmission, and species are the most inclusive units of evolutionary transformation. But there is no agreement on what the biological context is in which characters or homologues act as units. Homologues, if they are natural kinds, do not exist in order to serve the needs of comparative anatomists. There has to be a biological reason why the bodies of higher organisms are so obviously built in a modular way such that apparently natural units are often easy to recognize. In this paper it is argued that homologues can be understood as modular units of evolutionary transformation. In addition, the selection forces are discussed which may be responsible for the evolution of modularity. A way to test the suggested scenario is shortly outlined.