The 500-terminal (499-taxon) rbcL data set assembled and analyzed by Chase et al. in 1993, and re-analyzed by Rice et al. in 1997, is a useful benchmark for the comparative evaluation of available methods and software for phylogenetic analysis. Chase et al. detected trees of length 16225 (with uninformative characters deleted), and Rice et al. detected trees of length 16220. Trees two steps shorter (16218 steps) have been discovered numerous times using the parsimony ratchet, and these are likely to be the most-parsimonious trees for this matrix. We have analyzed this data set by conventional means (i.e., non-ratchet searches) to determine whether such searches can discover trees of length 16218 in reasonably short time periods, and if so, which settings are most efficient for the analysis of this data set. The results are also applicable to searches that use the ratchet and other methods, for they identify the factors that have substantial effects on the efficiency of tree searches whenever branch swapping is conducted. Trees of length 16218 were discovered by more than 150 conventional searches, and trees of length 16220 or shorter were discovered by more than 1500 conventional searches, representing a variety of strategies and settings. The most efficient searches involve multiple stages, with relatively few trees held during each stage.

Key words: angiosperms, cladistic analysis, large data sets, parsimony, rbcL