Leishmania infantum is the agent of visceral leishmaniasis in the Mediterranean basin. It is transmitted by sandflies of the subgenus Larroussius. Although Phlebotomus perniciosus is the most important vector in this area, an atypical Ph. perniciosus easily confused with Ph. longicuspis has been observed in North Africa. MALDI-TOF MS, an important tool for vector identification, has recently been applied for the identification of sandflies. Spectral databases presented in the literature, however, include only a limited number of Larroussius species. Our objective was to create an in-house database to identify Mediterranean sandflies and to evaluate the ability of MALDI-TOF MS to discriminate close species or atypical forms within the Larroussius subgenus. Field-caught specimens (n = 94) were identified morphologically as typical Ph. perniciosus (PN; n = 55), atypical Ph. perniciosus (PNA; n = 9), Ph. longicuspis (n = 9), Ph. ariasi (n = 9), Ph. mascittii (n = 3), Ph. neglectus (n = 5), Ph. perfiliewi (n = 1), Ph. similis (n = 9) and Ph. papatasi (n = 2). Identifications were confirmed by sequencing of the mtDNA CytB region and sixteen specimens were included in the in-house database. Blind assessment on 73 specimens (representing 1073 good quality spectra) showed a good agreement (98.5%) between MALDI-TOF MS and molecular identification. Discrepancies concerned confusions between Ph. perfiliewi and Ph. perniciosus. Hierarchical clustering did not allow classification of PN and PNA. The use of machine learning, however, allowed discernment between PN and PNA and between the lcus and lcx haplotypes of Ph. longicuspis (accuracy: 0.8938 with partial-least-square regression and random forest models). MALDI-TOF MS is a promising tool for the rapid and accurate identification of field-caught sandflies. The use of machine learning could allow to discriminate similar species.