An ensemble-trained chemical mass balance (CMB) approach is developed for particulate matter (PM) source apportionment (SA), particularly for use in health studies. The approach uses results from a short-term emission-based chemical transport model (CTM) and multiple receptor-based approaches. Ensemble results have less day-to-day variation in source impacts and fewer biases between observed and estimated PM2.5 mass compared to the original receptor model results. Ensemble results show increases in road dust, biomass burning, and coal impacts, but secondary organic carbon (SOC) impacts decrease. These results, along with observations, are then used to obtain new source profiles. Two sets of new source profiles based on ensemble results in summer (July 2001 and winter (January 2002) were developed, and used in separate CMB applications for a 12-month data set of daily PM2.5 measurements at the Atlanta, GA, Jefferson Street site. Results show that ensemble-trained CMB approaches, using both summer profiles and winter profiles, effectively reduce day-to-day variability of source impact estimates by reducing fewer days of zero impact from sources known to be present as compared to traditional receptor modeling, suggesting improved results.