As the weak absorption intensity and high interference signals in the thermal infrared band of carbon monoxide （CO）， it is difficult to retrieve CO profiles with promising accuracy from thermal infrared data. The development and application of ultra-spectral infrared detector make it possible to improve the retrieval accuracy of CO profile. However， the ultra-spectral resolution and the huge channel numbers of the data not only enhance the abundant atmospheric retrieval information， but also induce lots of redundant information. As such， it is necessary to do the channel selection to ensure the accuracy and efficiency of retrieval. In this paper， a channel selection method considering both channel sensitivity and weighting function characteristics is proposed to CO profile retrieval of ultra-spectral infrared data. First， by analyzing the gas sensitivity of the channels in CO absorption band， the channels severely affected by other gases are excluded and initial channel group is obtained. Then， the weighting function characteristics of the initial channel group are studied. The channels located at the bottom and top of the peaks at the CO absorption spectrum， suggesting abundant gas retrieval information about different atmospheric layer， are selected as the final channel selection results. The channel selection method is applied for the winter and summer air masses in Alxa desert area， Beijing-Tianjin area， Yangtze River Delta， and Pearl River Delta. By comparing with the Optimal Sensitivity Profile method （OSP）， the channels selected by the proposed method can cover a wider spectral range and have more CO absorption characteristics. Additionally， the application of the proposed method can improve the retrieval accuracy of CO profiles in all the regions and seasons studied in this paper. The best improvement effect was observed in the Alxa desert area in the winter， whose root mean square error （RMSE） was reduced from 3.23×10^-8 g/g to 3.07×10^-8 g/g， with an average increase in accuracy of 10.56%.