This study investigates species boundaries in the lichen genus Arctomia (Arctomiaceae, Ascomycota) using an integrative approach combining molecular phylogenetics, full Bayesian population delimitation, heuristic and model-based species delimitation, and supervised machine learning applied to morphological data. We analysed DNA sequence data from four markers: the internal transcribed spacer (ITS) region and the large subunit (nrLSU) of the ribosomal RNA gene, the DNA-directed RNA polymerase II subunit RPB1 (RPB1), and the small subunit of the mitochondrial ribosomal RNA (mrSSU) gene. Additionally, we assessed morphological measurements of ascospores and apothecia. Phylogenetic analyses consistently recovered Arctomia, also including A. interfixa, as monophyletic. The traditional classification of A. delicatula was found to encompass multiple species. We recognise three distinct species within this complex: A. delicatula s.str., A. acutior (resurrected from synonymy with A. delicatula), and a newly described species, A. confusa. Supervised machine learning, trained on morphological data from sequenced specimens, successfully learnt morphological differences between species in the training set and subsequently classified unseen (unsequenced) specimens, including type material, with high probabilities. Arctomia acutior can be distinguished from A. delicatula and A. confusa by its narrower ascospores, smaller apothecia, and preference for bark substrates. While A. confusa is similar to A. delicatula, it can be separated by slightly narrower ascospores. Arctomia acutior inhabits decaying bark, A. delicatula s.str. overgrows soil and bryophytes on soil, whereas A. confusa can be found in either of the habitats. Our study demonstrates that the novel application of machine learning trained on morphological measurements from sequenced specimens offers a promising tool for distinguishing between species when morphological differences are subtle and for classifying historical specimens (including types) based on morphology when DNA sequences are not available. Keywords Baeomycetales; coalescent; machine learning; phylogeny; speciation; taxonomy.