Rapid, ancient radiations pose one of the most difficult challenges for phylogenetic estimation. We used DNA sequence data of 9,006 aligned base pairs from five genes (chloroplast atpB, matK, rbcL, and 18S and 26S nrDNA) to elucidate relationships among major lineages of Saxifragales (angiosperms, eudicots). These relationships were poorly supported in previous studies, apparently because the lineages originated in rapid succession. Using an array of methods that explicitly incorporate assumptions about evolutionary process (weighted maximum parsimony, maximum likelihood, LogDet/paralinear transformed distances), we show that the initial diversification of Saxifragales was indeed rapid. We suggest that the poor resolution of our best phylogenetic estimate is not due to violations of assumptions or to combining data partitions having conflicting histories or processes. We show that estimated branch lengths during the initial diversification are exceedingly short, and we estimate that acquiring sufficient sequence data to resolve these relationships would require an extraordinary effort (≈107 bp), assuming a linear increase in branch support with branch length. However, our simulation of much larger data sets containing a distribution of phylogenetic signal similar to that of the five sampled gene sequences suggests a limit to achievable branch support. Using statistical tests of differences in the likelihoods of topologies, we evaluated whether the initial radiation of Saxifragales involved the simultaneous origin of major lineages. Our results are consistent with predictions that resolving the branching order of rapid, ancient radiations requires sampling characters that evolved rapidly at the time of the radiation but have since experienced a slower evolutionary rate.