Modeling of secondary organic aerosol (SOA) has remained a big challenge due to the various precursors and complex processes involved. In this study, the WRF-CAMx model was used to predict the ambient SOA concentrations in urban Beijing as well as the North China Plain (NCP) during a polluted period in winter. To identify the major uncertainties and improve the model performance, a series of model tests were performed to assess the sensitivity of model prediction to the key factors. Then the sources of SOA in Beijing were identified using the optimized model. Both the volatility basis set (VBS) approach and the two-product approach were used for SOA simulation. Although the modeled SOA was underpredicted compared with the SOA estimated through filter-based measurements, the VBS scheme produced higher SOA than the traditional two-product scheme. According to the sensitivity tests with the VBS scheme, the emissions of volatile organic compounds (VOC) and intermediate volatility organic compounds (IVOC) as well as the oxidant levels were the key factors that affected SOA prediction. Based on the optimized simulation scenario, the potential contributions from different anthropogenic sources and source areas were identified, with over 80% of SOA in urban Beijing from regional transport of SOA or its precursors from the surrounding areas during the polluted period. Residential emission in the North China Plain appeared as the dominant source of SOA in urban Beijing from the perspective of regional contribution.