We have used three reference sequences representative of bacterial drug resistance pumps and sugar transport proteins to collect the 91 most closely related sequences from a composite, nonredundant protein sequence database. Having eliminated certain very close relatives, the remainder were subjected to analysis and alignment by using two different similarity matrices: one of these was a matrix based on structural conservation of amino acid residues in proteins of known conformation and the other was based on the more familiar mutational matrix. Unrooted similarity trees for these proteins were constructed for each matrix and compared. A systematic analysis of the differences between these trees was undertaken and the sequences were analyzed for the presence or absence of certain sequence motifs. The results show that the clades created by the two methods are broadly comparable but that there are some clusters of sequences that are significantly different. Further analysis confirmed that (1) the sequences collected by this objective method are all known or putative 12-helix (in some cases reported as 14-helix) transmembrane proteins, (2) there is evidence for few cases of an origin based on gene duplication, (3) the bacterial drug resistance pumps are distributed in more than one clade and cannot be regarded as a definitive subset of these proteins, and that (4) the diversity is such that there is no evidence of a single ancestral protein. The possible extension of the methods to other cases of divergent protein sequences is discussed.