It is commonly assumed that closely related species share more similar niches than do distantly related species, thus limiting their ability to coexist and leading to patterns of phylogenetic over-dispersion. On the contrary, recent theoretical developments argue that competitive exclusion may lead to patterns of either over-dispersion, clustering or randomness, depending on the relative importance of niche differences and interspecific competitive ability differences. In this study, we utilized semi-natural grassland communities to test the hypothesis that the pattern of species relatedness generated by competitive exclusion depends on species niche differences. Instead of inferring processes from observed patterns, we experimentally manipulated grassland plots to test the effects of competitive exclusion. We compared grazed plots (in which grazing functioned as an equalizing mechanism and suppressed aboveground competition) with neighbouring plots experimentally excluded from grazing over the course of nine years (in which aboveground competition was not prevented). Weak niche differences between species allowed competitive exclusion to generate phylogenetic clustering, because phylogenetically structured plant canopy height and capacity for lateral spread conferred superior competitive ability to grasses that outcompeted species in all dicot branches. By contrast, moderate niche differences allowed competitive exclusion to result in a random pattern of phylogenetic species assembly, because species niche differences partially counterbalanced competitive exclusion by superior competitors in the grass family. We conclude that patterns of species relatedness created by competitive exclusion in Iberian Atlantic grasslands depend on species niche differences. Competitive exclusion may cause differential patterns of phylogenetic assembly.