Abstract. For PM2.5 filter samples collected daily at the Chinese Research Academy of Environmental Sciences (Beijing, China) from December of 2013 to February of 2014 (the winter period), chemical characteristics and sources were investigated with an emphasis on haze events in different alert levels. During the 3 months, the average PM2.5 concentration was 89 µg m−3, exceeding the Chinese national standard of 75 µg m−3 in 24 h. The maximum PM2.5 concentration was 307 µg m−3, which characterizes developed-type pollution (PM2.5 / PM10>0.5) in the World Health Organization criteria. PM2.5 was dominated by SO42−, NO3−, and pseudo-carbonaceous compounds with obvious differences in concentrations and proportions between non-haze and haze episodes. The non-negative matrix factorization (NMF) analysis provided reasonable PM2.5 source profiles, by which five sources were identified: soil dust, traffic emission, biomass combustion, industrial emission, and coal combustion accounting for 13, 22, 12, 28, and 25 % of the total, respectively. The dust impact increased with northwesterlies during non-haze periods and decreased under stagnant conditions during haze periods. A blue alert of heavy air pollution was characterized by the greatest contribution from industrial emissions (61 %). During the Chinese Lantern Festival, an orange alert was issued and biomass combustion was found to be the major source owing to firework explosions. Red-alert haze was almost equally contributed by local traffic and transported coal combustion emissions from the vicinity of Beijing (approximately 40 % each) that was distinguished by the highest levels of NO3− and SO42−, respectively. This study also reveals that the severity and source of haze are largely dependent on meteorological conditions.