The multivariate receptor models Positive Matrix Factorization (PMF) and Unmix were used along with the EPA's Chemical Mass Balance model to deduce the sources of PM2.5 at a centrally located urban site in Seattle, WA. A total of 289 filter samples were obtained with an IMPROVE sampler from 1996 through 1999 and were analyzed for 31 particulate elements including temperature-resolved fractions of the particulate organic and elemental carbon. All three receptor models predicted that the major sources of PM2.5 were vegetative burning (including wood stoves), mobile sources, and secondary particle formation with lesser contributions from resuspended soil and sea spray. The PMF and Unmix models were able to resolve a fuel oil combustion source as well as distinguish between diesel emissions and other mobile sources. In addition, the average source contribution estimates via PMF and Unmix agreed well with an existing emissions inventory. Using the temperature-resolved organic and elemental carbon fractions provided in the IMPROVE protocol, rather than the total organic and elemental carbon, allowed the Unmix model to separate diesel from other mobile sources. The PMF model was able to do this without the additional carbon species, relying on selected trace elements to distinguish the various combustion sources.