This paper introduces the basic theory and method of carbon dioxide (CO2) retrieval. The key step is to search for the optimal solution and the random search algorithm Genetic Algorithm (GA) which can effectively avoid the local optimization. We first investigate the basic principles of GA in CO<sub>2</sub> retrieval and then design the corresponding encoding and decoding methods as well as the fitness function. This newly-developed GA is further applied to retrieve the atmospheric CO<sub>2</sub> concentration using Atmospheric Infrared Sounder (AIRS) observations from January 2006 to December 2008 centered at 20°N, 144°E. Compared to the aircraft measurements, the GA retrieval yields the small root mean square error of 1.13 ppmv and reproduces good results with the observed seasonal cycle.