On- orbit radiometric calibration data from some types of atmospheric radiation measurement satellite are analyzed to investigate the impact of pollutants on the response performance of space- based infrared camera. An attenuation model of camera responded grayscale is established with a clear physical meaning, whose parameters are fitted using the practical data with the relative root mean square errors less than 1% , and the fitted values coincide with their physical meanings well. In order to determine the pollution tolerance in satellite, rules based on the constraints of camera linear response range and the minimum radiance resolution requirements are proposed, and the recommended decontamination interval of 1250 h is obtained. The method of using the attenuation model to predict calibration coefficients at any moment is presented, and calibration coefficients from model prediction, history calibration tests and the time- matched calibration test are used to convert measuring grayscale data to radiance. Results show that the modelpredicted coefficients can control the calibration errors with less than 3% and are able to improve the measurement accuracy while ensuring image quality.