Evolutionary interpretation of paleontological patterns requires a hypothes
is of phylogeny, but our phylogenetic hypotheses may not perfectly mirror o
rganismal phylogeny. Tree summary methods less conservative than strict con
sensus may increase resolution, but these methods may present a biased summ
ary of the full set of most parsimonious trees. When we fail to ac knowledg
e all equally optimal topologies, we risk disregarding trees that are close
r to the correct phylogeny. We discuss a case where two subsets of trees we
re recovered in the set of most parsimonious trees, each with a profoundly
different interpretation of character evolution near the root of Echinoderm
ata. This was caused by the presence of a bimodally labile taxon in the mat
rix with two different topological subsets, each equally parsimonious but d
iffering in the number of consistent trees. Majority-rule consensus favors
the subset with the largest number of trees consistent with the placement o
f the rogue taxon. This bias favors clusters not because of the biological
implications of the tree, but on the basis of great inequality in the sizes
of the islands of parsimony. We thus recommend that majority-rule consensu
s trees not be used to summarize the results of a phylogenetic analysis.