AbstractPhylogenetic analysis algorithms require the assumption of character independence - a condition generally acknowledged to be violated by morphological data. Correlation between characters can originate from intra-organismal features, shared phylogenetic history or forced by particular character-state coding schemes. Although the two first sources can be investigated by biologists a posteriori and the third one can be avoided a priori with good practices, phylogenetic software do not distinguish between any of them.In this study, we propose a new metric of raw character difference as a proxy for character correlation. Using thorough simulations, we test the effect of increasing or decreasing character differences on tree topology. Overall, we found an expected positive effect of reducing character correlations on recovering the correct topology. However, this effect is less important for matrices with a small number of taxa (25 in our simulations) where reducing character correlation is not more effective than randomly drawing characters. Furthermore, in bigger matrices (350 characters), there is a strong effect of the inference method with Bayesian trees being consistently less affected by character correlation than maximum parsimony trees.These results suggest that ignoring the problem of character correlation or independence can often impact topology in phylogenetic analysis. However, encouragingly, they also suggest that, unless correlation is actively maximised or minimised, probabilistic methods can easily accommodate for a random correlation between characters.