Gene tree discordance is now recognized as a major source of biological heterogeneity. How to deal with this heterogeneity is an unsolved problem, as the accurate inference of individual gene tree topologies is difficult. One solution has been to simply concatenate all of the data together, ignoring the underlying heterogeneity. Another approach infers gene tree topologies separately and combines the individual estimates in order to explicitly model this heterogeneity. Here we discuss the advantages and disadvantages of both approaches-using the gene trees singly or in concatenation-paying special attention to the sources of variance and implicit assumptions. We make it clear that all methods are likely to have their assumptions violated, though the consequence of these violations differs in different parts of parameter space. The main conclusion of our review is that different sources of error are more or less important in different settings, such that phylogenetics researchers should be using the methods most appropriate to their problems rather than stick to one dogmatically.