DNA barcoding, being developed for biomonitoring, requires a database of reference sequences and knowledge of how much sequences can deviate before they are assigned to separate species. The molecular hunt for hidden species also raises the question of species definitions. We examined whether there are objective criteria for sequence-based species delimitation in diatoms, using Nitzschia palea, an important monophyletic indicator species already known to contain cryptic diversity. Strains from a wide geographical range were sequenced for 28S rRNA, COI and rbcL. Homogeneity indices and the Chao index failed to objectively select a precise number of species existing in N. palea as well as an evolutionary method based on coalescence theory. COI always gave higher diversity estimations than 28S rRNA or rbcL. Mating data did not provide a precise calibration of molecular species thresholds. Rarefaction curves indicated that further MOTUs would be detected with more isolates than we sampled (81 clones, 42 localities). Although some genotypes had intercontinental distributions, there was a positive relationship between genetic and geographical distance, suggesting even higher richness than we assessed, given that many regions were not sampled. Overall, no objective criteria were found for species separation; instead barcoding will need a consensual approach to molecular species limits.