Since earlier investigations gave conflicting results, we re-examined the phylogeny of the Lauraceae with an enlarged data set from the trnK intron, including a higher number of critical taxa and a larger number of coding (matK) and non-coding sequence positions. The source of earlier conflicts was explored by using different algorithms (parsimony, maximum likelihood, and Bayesian inference), different taxon samples (by exclusion of one or two of the critical taxa), and different parts of the matrix (entire trnK intron, coding or non-coding positions only). The results suggest that (1) Bayesian inference (and in smaller data sets also maximum likelihood) yields results that are more consistent with the distribution of informative indels than the results of parsimony analyses. (2) the monotypic African genus Hypodaphnis is sister to all other Lauraceae. (3) Neocinnamomum is indeed close to Caryodaphnopsis, as morphology would suggest. and (4) the parasitic genus Cassytha is not sub-basal, as in our earlier analysis, but nested among the otherwise woody Lauraceae as sister group to a clade including all genera except Hypodaphnis and the Cryptocarya group.