To most evolutionary biologists, building phylogenetic trees is a bore. Comparative approaches to understanding patterns as diverse as the correlation of body size and brain size in primates, and the evolution of extended phenotypes in gallwasps, require a robust phylogeny on which to place changes in character state. But, for every question you are really interested in there are a thousand petty arguments concerning tree building. Which method should you use? Which parameter values should you plug in? How much can you trust the results? As Joe Felsenstein wrote, ‘Comparative biologists may understandably feel frustrated upon being told that they need to know the phylogenies of their groups in great detail, when this is not something they had much interest in knowing.’ However, he also goes on to say ‘Nevertheless phylogenies are fundamental to comparative biology; there is no doing it without taking them into account.’ Phylogenetic trees might be a nuisance, but until now you could not do without them. And, if your ‘favourite’ tree was wrong, so were your conclusions. However, a recent paper by Huelsenbeck et al.1xAccommodating phylogenetic uncertainty in evolutionary studies. Huelsenbeck, J.P. et al. Science. 2000; 288: 2349–2350Crossref | PubMed | Scopus (203)See all References1 has raised an exciting new possibility: evolution without trees.The authors do not suggest that you don’t do your phylogenetic tree-building homework, but that you can eliminate the uncertainty involved in tree building by considering all possible phylogenies and parameter values, weighting according to the probability that each is correct. The idea is just one more application of a field that is becoming increasingly common in evolutionary analysis: Bayesian statistics. In practice, this means that you replace parameters, such as tree topologies, biases in DNA mutation and rates of change of characters, by distributions representing your certainty, or otherwise, about them. These can be based on previous analyses, current data or even biological intuition. The probability of a given hypothesis being correct given the observed data is then summed over all possible realizations, weighting each by their relative probability. As Huelsenbeck et al.1xAccommodating phylogenetic uncertainty in evolutionary studies. Huelsenbeck, J.P. et al. Science. 2000; 288: 2349–2350Crossref | PubMed | Scopus (203)See all References1 demonstrate, this can lead to considerably different conclusions from what popular methods of tree reconstruction, such as parsimony, would suggest. More importantly, it gives you a good feel for how much faith to have in your conclusions.Performing this task requires some serious computational effort, but there is new software appearing that can actually do the number crunching in reasonable time. Perhaps, more importantly, Bayesian analysis relies exclusively on the likelihood framework for tree building, and die-hard opponents of such model-based methods could argue that unless you are sure that the model of sequence evolution you are using is correct, the likelihood value is meaningless. Of course, you can always include more parameters, but this is a path that can spiral rapidly into absurdity. Nevertheless, tree felling the Bayesian way is going to make a large difference to everyone working with the comparative method because it eliminates the potential nightmare of using the wrong tree. As the lumberjacks say, if you can’t see the wood for the trees, chop them down.