The process of discovering species is a fundamental responsibility of systematics. Recently, there has been a growing interest in coalescent-based methods of species delimitation aimed at objectively identifying species early in the divergence process. However, few empirical studies have compared these new methods with character-based approaches for discovering species. In this study, we applied both a character-based and a coalescent-based approaches to delimit species in a closely related avian complex, the light-vented/Taiwan bulbul (Pycnonotus sinensis/Pycnonotus taivanus). Population aggregation analyses of plumage, mitochondrial and 13 nuclear intron character data sets produced conflicting species hypotheses with plumage data suggesting three species, mitochondrial data suggesting two species, and nuclear intron data suggesting one species. Such conflict is expected among recently diverged species, and by integrating all sources of data, we delimited three species verified with independently congruent character evidence as well as a more weakly supported fourth species identified by a single character. Attempts to validate species hypothesis using Bayesian Phylogenetics and Phylogeography (BPP), a coalescent-based method of species delimitation, revealed several issues that can seemingly affect statistical support for species recognition. We found that θ priors had a dramatic impact on speciation probabilities, with lower values consistently favouring splitting and higher values consistently favouring lumping. More resolved guide trees also resulted in overall higher speciation probabilities. Finally, we found suggestive evidence that BPP is sensitive to the divergent effects of nonrandom mating caused by intraspecific processes such as isolation-with-distance, and therefore, BPP may not be a conservative method for delimiting independently evolving population lineages. Based on these concerns, we questioned the reliability of BPP results and based our conclusions about species limits exclusively on character data.