As one of the main greenhouse gases, global atmospheric carbon dioxide has increased one third over the past one hundred years, becoming the focus of global warming research. Using satellite remote sensing technology and radiative transfer model to retrieve atmospheric carbon dioxide has become the most important means to obtain the global carbon dioxide concentration data. However, the input parameters of radiative transfer model inevitably carry some errors, which will influence the retrieval accuracy more or less, so this influence must be quantified and considered in the retrieval algorithm. In this paper, the radiances of 17 infrared carbon dioxide retrieval channels located in 15mum wave band of the Aqua/AIRS infrared detection instrument are simulated by the fast radiative transfer model RTTOV10. The temperature profile, ozone profile, water vapor profile, surface temperature and emissivity are considered as the influence factors. And the sensitivities of the outgoing radiance to the errors of these input parameters are calculated. Then the comparison between these sensitivities and that to 0.5% change of atmospheric carbon dioxide concentration is made. Analysis shows that: temperature profile errors have the biggest influence on carbon dioxide retrieval, and these errors should be less than 0.5k. The total sensitivity of all channels to ozone profile errors are about one third of that to carbon dioxide, making ozone profile errors take the second place. The sensitivities to errors of water vapor profile and surface temperature show the same trend which is only high at channel 738.1 cm~(-1) and 738.4 cm~(-1). Errors of surface emissivity are negligible at all channels, because the sensitivities due to them are far less than that to carbon dioxide. Finally, for each channel, these parameters are divided into three classes which are high sensitive parameters, low sensitive parameters and non-sensitive parameters, providing a reference for carbon dioxide retrieval channel selection and algorithm design.