Phylogenetic methods are key to providing models for how a given protein family evolved. However, these methods run into difficulties when sequence divergence is either too low or too high. Here, we provide a case study of Hox and ParaHox proteins so that additional insights can be gained using a new computational approach to help solve old classification problems. For two (Gsx and Cdx) out of three ParaHox proteins the assignments differ between the currently most established view and four alternative scenarios. We use a non-phylogenetic, pairwise-sequence-similarity-based method to assess which of the previous predictions, if any, are best supported by the sequence-similarity relationships between Hox and ParaHox proteins. The overall sequence-similarities show Gsx to be most similar to Hox2–3, and Cdx to be most similar to Hox4–8. The results indicate that a purely pairwise-sequence-similarity-based approach can provide additional information not only when phylogenetic inference methods have insufficient information to provide reliable classifications (as was shown previously for central Hox proteins), but also when the sequence variation is so high that the resulting phylogenetic reconstructions are likely plagued by long-branch-attraction artifacts.