BackgroundMatrix-assisted laser desorption ionization-time of flight mass spectrometry (MALDI-TOF MS) for yeast identification is limited by the requirement for protein extraction and for robust reference spectra across yeast species in databases. We evaluated its ability to identify a range of yeasts in comparison with phenotypic methods.MethodsMALDI-TOF MS was performed on 30 reference and 167 clinical isolates followed by prospective examination of 67 clinical strains in parallel with biochemical testing (total n = 264). Discordant/unreliable identifications were resolved by sequencing of the internal transcribed spacer region of the rRNA gene cluster.Principal FindingsTwenty (67%; 16 species), and 24 (80%) of 30 reference strains were identified to species, (spectral score ≥2.0) and genus (score ≥1.70)-level, respectively. Of clinical isolates, 140/167 (84%) strains were correctly identified with scores of ≥2.0 and 160/167 (96%) with scores of ≥1.70; amongst Candida spp. (n = 148), correct species assignment at scores of ≥2.0, and ≥1.70 was obtained for 86% and 96% isolates, respectively (vs. 76.4% by biochemical methods). Prospectively, species-level identification was achieved for 79% of isolates, whilst 91% and 94% of strains yielded scores of ≥1.90 and ≥1.70, respectively (100% isolates identified by biochemical methods). All test scores of 1.70–1.90 provided correct species assignment despite being identified to “genus-level”. MALDI-TOF MS identified uncommon Candida spp., differentiated Candida parapsilosis from C. orthopsilosis and C. metapsilosis and distinguished between C. glabrata, C. nivariensis and C. bracarensis. Yeasts with scores of <1.70 were rare species such as C. nivariensis (3/10 strains) and C. bracarensis (n = 1) but included 4/12 Cryptococcus neoformans. There were no misidentifications. Four novel species-specific spectra were obtained. Protein extraction was essential for reliable results.ConclusionsMALDI-TOF MS enabled rapid, reliable identification of clinically-important yeasts. The addition of spectra to databases and reduction in identification scores required for species-level identification may improve its utility.