Catfishes (Teleostei: Siluriformes) are a highly diverse order within Ostariophysi that is distributed worldwide. At the base of this clade emerge three lineages with well-defined monophylies: Diplomystidae, Loricarioidei, and Siluroidei. Morphological phylogeny studies place the Diplomystidae as the earliest branching of these three lineages, but studies based on molecular phylogenetics consistently find the fast-evolving Loricarioidei instead. The high lineage evolutionary rate heterogeneity in this order and the fact that the lineage placed closest to the root in the molecular phylogenies is fast evolving, including many long branches, raises the possibility that the discrepancy between morphological and molecular phylogenies may be the result of a long branch attraction inference artifact. We test this hypothesis by using a 10-gene dataset to evaluate the arrangement of the three main siluriform lineages, and apply the LS3 and LS4 taxon sequence subsampling methods to reduce evolutionary rate heterogeneity among lineages. The initial and complete dataset supports the basal branching of Loricarioidei as in all previous molecular phylogenies, but once lineage rate heterogeneity is reduced with LS3 or LS4 through the removal of sequences disrupting homogeneity, the phylogeny shows Diplomystidae as the earliest branching group, with high supports, as proposed by morphological phylogeny. The result obtained with LS3, however, introduces the misplacement of one of the species with the highest amount of missing data, Scoloplax sp. Because the sequence sub-selection criterion of LS4 has been optimized to reduce data removal, the phylogeny resulting from the LS4-processed data is in agreement with the known intra-lineage relationships in addition to supporting the morphologically-based rooting hypothesis. Our results are the first instance in which a consensus between molecular and morphological phylogeny is reached concerning the root of this order.