Cloud contamination of greenhouse gas satellite data is investigated by using O2-A band, the scattering effects on CO2 retrieval from optical path-length is analyzed and a new method for solving cloud contamination problem is proposed. The results show that 83.58%, 9.70%, and 6.72% L2 data are contained in clear, cloud, and undetermined scenes. The CO2 retrieval of these clear scene and cirrus scene appears highly agreement with L2 product, but the other CO2 retrieval is generally lower than L2 product. Cloud screening in O2-A band has its own advantage but also existing insufficiency like the misidentifying of thin cirrus. Fortunately, the scattering correction method through optical path-length is an effective supplement to O2-A band cloud screening as it can improve CO2 retrieval bias due to cirrus scattering, combining them to account for cloud contamination is a promising method. Simultaneously, undetermined-I scene should be pay more attention since these undetermined scene contained abundance scene covered by cirrus.