The Aerosol Time-of-Flight Mass Spectrometer (ATOFMS) is one of few instruments able to measure the size and mass spectra of individual airborne particles with high temporal resolution. Data analysis is challenging and in the present study, we apply three different techniques (PMF, ART-2a and K-means) to a regional ATOFMS dataset collected at Harwell, UK. For the first time, Positive Matrix Factorization (PMF) was directly applied to single particle mass spectra as opposed to clusters already generated by the other methods. The analysis was performed on a total of 56,898 single particle mass spectra allowing the extraction of 10 factors, their temporal trends and size distributions, named CNO-COOH (cyanide, oxidized organic nitrogen and carboxylic acids), SUL (sulphate), NH4-OOA (ammonium and oxidized organic aerosol), NaCl, EC+ (elemental carbon positive fragments), OC-Arom (aromatic organic carbon), EC- (elemental carbon negative fragments), K (potassium), NIT (nitrate) and OC-CHNO (organic nitrogen). The 10 factor solution from single particle PMF analysis explained 45% of variance of the total dataset, but the factors are well defined from a chemical point of view. Different EC and OC components were separated: fresh EC (factor EC-) from aged EC (factor EC+) and different organic families (factors NH4-OOA, OC-Arom, OC-CHNO and CNO-COOH). A comparison was conducted between PMF, K-means cluster analysis and the ART-2a artificial neural network. K-means and ART-2a give broadly overlapping results (with 9 clusters, each describing the full composition of a particle type), while PMF, by effecting spectral deconvolution, was able to extract and separate the different chemical species contributing to particles, but loses some information on internal mixing. Relationships were also examined between the estimated volumes of ATOFMS PMF factors and species concentrations measured independently by GRAEGOR and AMS instruments, showing generally moderate to strong correlations.