Frequent occurring of haze pollution events and high fine particulate matter (PM2.5) concentration in China have attracted more and more attention in the world. Satellite remote sensing can be used to characterize the air pollution. However, haze is usually misidentified as fog, thin cloud or bright surface in NASA’s Moderate Resolution Imaging Spectrometer (MODIS) cloud and clear days’ aerosol products, and the retrieval of its optical properties is not included in MODIS cloud detection and dark target algorithm. This approach first studies the spectral characters of cloud, fog, haze, and land cover pixels. Second, following the previous cloud detection and aerosol retrieval literatures, a threshold algorithm is developed to distinguish haze from other pixels based on MODIS multi-band apparent reflectance and brightness temperature. This algorithm is used to detect the haze distribution over North China Plain in 2008 spring and summer. Our result shows a good agreement with the true-color satellite images, which enhances MODIS’s ability to monitor the severe air pollution episodes. In addition, the high AOD data from Beijing and Xiang Aerosol Robotic NETwork (AERONET) sites indicate nearly 80% haze days are detected by our approach. Finally, we analyze the errors and uncertainties in haze detection algorithm, and put forward the potential improvements.