This study describes the atmospheric aerosol load encountered during the large-scale pollution episode that occurred in August 2003, by means of the aerosol optical thicknesses (AOTs) measured at 865 nm by the Polarization and Directionality of the Earth's Reflectances (POLDER) sensor and the simulation by the CHIMERE chemistry-transport model. During this period many processes (stagnation, photochemistry, forest fires) led to unusually high particle concentrations and optical thicknesses. The observed/simulated AOT comparison helps understanding the ability of the model to reproduce most of the gross AOT features observed in satellite data, with a general agreement within a factor 2 and correlations in the 0.4–0.6 range. However some important aerosol features are missed when using regular anthropogenic sources. Additional simulations including emissions and high-altitude transport of smoke from wildfires that occurred in Portugal indicate that these processes could dominate the AOT signal in some areas. Our results also highlight the difficulties of comparing simulated and POLDER-derived AOTs due to large uncertainties in both cases. Observed AOT values are significantly lower than the simulated ones (30–50%). Their comparison with the ground-based Sun photometer Aerosol Robotic Network (AERONET) measurements suggests, for the European sites considered here, an underestimation of POLDER-derived aerosol levels with a factor between 1 and 2. AERONET AOTs compare better with simulations (no particular bias) than POLDER AOTs.