Atmospheric aerosol is one of the most important factors that affect air quality of urban environment, meanwhile, it has important effects on human health. Traditional aerosol optical depth (AOD) retrieval algorithms are always suitable for dark areas with low surface reflectance including ocean and densely vegetated areas, however, for bright urban areas, surface reflectance is high and difficult to be determined, leading to great challenges. Aiming at this problem, a new improved approach of surface reflectance estimation is proposed and the underlying surfaces are divided into dark and bright areas. Surface reflectance is determined using the simulated relationships between the surface reflectance between visible and short-wave infrared channels and a priori surface reflectance dataset constructed with long time series of MODIS apparent reflectance images using the minimum value synthesis technology. Then aerosol retrieval is performed based on the radiative transfer theory with pre-calculated Look-up Tables. Beijing, which features complex surfaces and serious air pollution, is selected as the study area and the proposed algorithm is applied to the MODIS data for aerosol retrieval experiments. Four AErosol RObotic NETwork (AERONET) AOD ground-measured stations, Beijing, Xianghe, Beijing_CAMS and Beijing_RADI, and operational MODIS aerosol product (MOD04) are selected for validation and comparison purposes. Results showed that AOD retrievals are highly consistent with AERONET AOD ground measurements (R2=0.902) and showed overall higher accuracy with more detailed spatial distribution compared to MOD04 AOD products over bright urban areas.