Despite the increasing importance of molecular sequence data, morphology still makes an important contribution to resolving the phylogeny of many groups, and is the only source of data for most fossils. Most systematists sample morphological characters as broadly as possible on the principle of total evidence. However, it is not uncommon for sampling to be focussed on particular aspects of anatomy, either because characters therein are believed to be more informative, or because preservation biases restrict what is available. Empirically, the optimal trees from partitions of morphological data sets often represent significantly different hypotheses of relationships. Previous work on hard-part versus soft-part characters across animal phyla revealed significant differences in about a half of sampled studies. Similarly, studies of the craniodental versus postcranial characters of vertebrates revealed significantly different trees in about one third of cases, with the highest rates observed in non-avian dinosaurs. We test whether this is a generality here with a much larger sample of 81 published data matrices across all major dinosaur groups. Using the incongruence length difference (ILD) test and two variants of the incongruence relationship difference (IRD) test, we found significant incongruence in about 50% of cases. Incongruence is not uniformly distributed across major dinosaur clades, being highest (63%) in Theropoda and lowest (25%) in Thyreophora. As in previous studies, our partition tests show some sensitivity to matrix dimensions and the amount and distribution of missing entries. Levels of homomplasy and retained synapomorphy are similar between partitions, such that incongruence must partly reflect differences in patterns of homoplasy between partitions, which may itself be a function of modularity and mosaic evolution. Finally, we implement new tests to determine which partition yields trees most similar to those from the entire matrix. Despite no bias across dinosaurs overall, there are striking differences between major groups. The craniodental characters of Ornithischia and the postcranial characters of Saurischia yield trees most similar to the ‘total evidence’ trees derived from the entire matrix. Trees from these same character partitions also tend to be most stratigraphically congruent: a mutual consilience suggesting that those partitions yield more accurate trees.