Despite extensive phylogenetic analysis of small subunit ribosomal RNA (SSUrRNA) genes, the deep-level relationships among kinetoplastids remain poorly understood, limiting our grasp of their evolutionary history, especially the origins of their bizarre mitochondrial genome organizations. In this study we examine the SSUrRNA data in the light of a new marker--cytoplasmic heat shock protein 90 (hsp90) sequences. Our phylogenetic analyses divide kinetoplastids into four main clades. Clades 1-3 include the various bodonid kinetoplastids. Trypanosomatids comprise the fourth clade. SSUrRNA analyses give vastly different and poorly supported positions for the root of the kinetoplastid tree, depending on the out-group and analysis method. This is probably due to the extraordinary length of the branch between kinetoplastids and any out-group. In contrast, almost all hsp90 analyses place the root between clade 1 (including Dimastigella, Rhynchomonas, several Bodo spp., and probably Rhynchobodo) and all other kinetoplastids. Maximum likelihood and maximum likelihood distance analyses of hsp90 protein and second codon-position nucleotides place trypanosomatids adjacent to Bodo saltans and Bodo cf. uncinatus (clade 3), as (weakly) do SSUrRNA analyses. Hsp90 first codon- plus second codon-position nucleotide analyses return a slightly different topology. We show that this may be an artifact caused, in part, by the different evolutionary behavior of first- and second-codon positions. This study provides the most robust evidence to date that trypanosomatids are descended from within bodonids and that B. saltans is a close relative of trypanosomatids. A total reevaluation of the high-level systematics within kinetoplastids is needed. We confirm that the interlocking network organization of kinetoplast DNA seen in trypanosomatids is a derived condition within kinetoplastids but suggest that open-conformation minicircles may have arisen early in kinetoplastid evolution. Further understanding of the evolution of kinetoplast structure and RNA editing is hampered by a paucity of data from basal (i.e., clade 1) bodonids.