The conditional probability of reconstruction is a measure of the robustness of cladogram internodes and, unlike Bremer support and bootstrapping values, directly gauges probability. The new method compares the three putative branch lengths (the optimal and two alternatives) obtained through branch recalculation after nearest neighbor interchange and recalculation under constraint. With rooted trees, one switches the two free lineages attached at the distal end of an internal branch with the

more »

... al lineage. Probabilistic reconstruction of a branch for small data sets (e.g., morphological) is de ned as having no contrary support for the two alternative branches and, when suf cient data are available (e.g., molecular studies), as meeting a selected con dence limit in chi-squared analysis. The exact probability that the internal branch is reconstructed is the same as the preselected con dence level met with chi-squared analysis; alternatively, it is a simple calculation of the length of the optimal branch divided by the sum of the lengths of all three putative branches. This new measure of robustness allows calculation of summary probabilities of subclade and tree reconstruction. The measure is conditional on a particular data set and optimization method but may also compare support from con icting gene trees. Examples are provided by a morphological data set (the bryophyte Didymodon) and a molecular data set (