Samples of PM 10 and PM 2.5 were synchronously collected from a megacity in China (Chengdu) during the 2011 sampling campaign and then analyzed by an improved three-way factor analysis method based on ME2 (multilinear engine 2), to investigate the contributions and size distributions of the source categories for size segregated particulate matter (PM). Firstly, the synthetic test was performed to evaluate the accuracy of the improved three-way model. The same five source categories with slightly different source profiles were caught. The low AAE (average absolute error) values between the estimated and the synthetic source contributions (<15%) and the approachable estimated PM 2.5 /PM 10 ratios with the simulated ratios might indicate that the results of the improved three-way factor analysis might be satisfactory. Then, for the ambient PM samples, the mean levels were 206.65±69.90μg/m 3 (PM 10 ) and 130.47±43.67μg/m 3 (PM 2.5 ). The average ratio of PM 2.5 /PM 10 was 0.63. PM 10 and PM 2.5 in Chengdu were influenced by the same source categories and their percentage contributions were in the same order: crustal dust & coal combustion presented the highest percentage contributions, accounting for 58.20% (PM 10 ) and 53.73% (PM 2.5 ); followed by vehicle exhaust & secondary organic carbon (18.45% for PM 10 and 21.63% for PM 2.5 ), secondary sulfate and nitrate (17.06% for PM 10 and 20.91% for PM 2.5 ) and cement dust (6.30% for PM 10 and 3.73% for PM 2.5 ). The source profiles and contributions presented slightly different distributions for PM 10 and PM 2.5 , which could better reflect the actual situation. The findings based on the improved three-way factor analysis method may provide clear and deep insights into the sources of synchronously size-resolved PM.