There's an article in the Sept. 28 issue of Nature discussing the regulation of mating type in two yeast species, c. albicans and s. cerevisiae (regretably, the text is only available if you're on a university network with access to the Nature archives). I'll spare you the gory details, but while the two yeast share the same observable trait (phenotype), the underlying genetic mechanism is completely different.
What is most interesting to me about the article is the detail in which a hypothesis is presented for the evolution of both systems from a common ancestor. The authors list the primary modifications necessary to change one trait into another, note that only a few mutations in one particular protein binding site would be needed to effect one of the modifications, point out that a third yeast species seems to represent an intermediate step between the two, and even suggest future work which includes engineering a strain of s. cerevisiae with the c. albicans mating system, to study what the ancestral yeast might have looked like.
This hypothesis is not only detailed but phenomenally testable (ironically, it involves neutral evolution with no net advantage to the organism). Compare this to the evolutionary arguments for the origin of the bacterial flagella, which rely exclusively on distant (and some not-so-distant) homologies and a good deal of story-telling and hand-waving.
The difference is immediately clear. If all evolutionary inferences were held to the standard of proof present in this nature article, I would have no objections. What I object to is the assumption that natural selection acting on random variation must be sufficiently poweful to fill the gaps between whatever homologies we happen to find in nature.