It has been thought that higher-level taxonomic phylogenetic studies focusing on ancient cladogenesis events would be better recovered using slow-evolving genes, thereby avoiding multiple hits and homoplasies. Furthemore, the distinction of "slow versus fast" has long been confusing: lower numbers of variable sites and lower frequency of change at variable sites being have both been used when referring to slowly-evolving genes. Hillis has addressed both issues using simulations of a 232-taxon 18S rDNA tree, and showed that rapidly-evolving characters should perform better in phylogeny reconstruction, providing that enough variable sites are used. Using a six-gene data set (atpB, atp1, rbcL, matR, matK, and 18S rDNA, from mitochondrial, plastid and nuclear genomes) for basal angiosperms, we evaluated Hillis' expectations and reached the same conclusion. Rapidly-evolving genes (i.e. those with more variable sites) performed the best (e.g. matK), indicating that large-scale phylogenetic analyses of plants will benefit from sequencing "fast" genes. Contrary to what is generally assumed, none of these regions differs much with respect to the frequencies of change at variable sites (i.e. variable positions are all evolving at similar rates).

Key words: fast versus slow evolving genes, Large-scale phylogeny