For 28 genera, including all genera (except Hesperomeles) and major generic segregates of Maloideae, we obtained about 7500 bp of sequences from four chloroplast DNA regions: the atpB-rbcL intergenic spacer, trnK intron plus matK gene, trnT-L-F region, and rps16 intron. For outgroups we used Kageneckia, Lindleya, and Vauquelinia, the closest relatives of the traditionally circumscribed, pome-fruited subfamily. Parsimony analyses of cpDNA data conflict strongly with nuclear data with respect to segregates of Sorbus. CpDNA supports (Chamaemespilus, Torminalis) at a 100% bootstrap (BS) level and (Aria, Cormus) at 71% BS in a larger, weakly supported clade that also includes Sorbus s.s. Results from the four granule-bound starch synthase genes, in contrast, have BS values of 68, 86, 89, and 90% for (Aria, Chamaemespilus), 81-96% BS for (Cormus, Sorbus s.s.), and no support for a relationship between these genera and Torminalis. CpDna and nuclear trees do agree on a crataegoid clade - ((Crataegus, Mespilus) (Malacomeles (Amelanchier, Peraphyllum)) - and (Eriobotrya, Rhaphiolepis). Otherwise cpDNA and nuclear trees do not resolve many deep nodes on their trees. This lack of resolution is at least partly the result of low sequence divergence (about 0.5-2.0% in cpDNA data), which is remarkable in that several maloid genera are known as fossils from the Middle Eocene. Gene flow between genera in the past, as suggested by the conflicts between cpDNA and nuclear data noted above, may explain some of this lack of evolutionary divergence. Hybridization has been reported between many genera of Maloideae, including, for example, all segregates of Sorbus s.l. except Cormus. The pattern of short internal branch lengths that we have reported for nuclear data also characterizes the most parsimonious cpDNA trees, frustrating efforts to understand higher-level relationships in the subfamily.

Key words: hybridization, Maloideae, phylogeny, Rosaceae