Urban Toronto fine particulate matter (PM2.5) was physically and chemically characterized by online aerosol laser ablation mass spectrometry (LAMS) between January 2002 and February 2003. The mass spectra from the analysis of individual aerosol particles were classified according to chemical composition by a neural network approach called adaptive resonance theory (ART-2a). Temporal trends of the hourly analysis rate of over 120 different particles types were constructed and subjected to positive matrix factorization (PMF). This receptor modeling technique enabled the identification of nine distinct emission sources responsible for these particle types: biogenic, mixed crustal, organic nitrate, construction dust, Toronto soil/road salt, secondary salt, wood burning, intercontinental dust, and an unknown source of aluminum fluoride dust. Episodic events occurred with the wood burning, intercontinental dust, and unknown dust sources. This is the first paper reporting the application of PMF to single-particle spectral data.