Abstract: To meet the demand for accurate atmospheric carbon dioxide (CO?) concentration data in regional carbon cycle research and emission reduction policy formulation, and to address the biases of multi-source datasets in characterizing the spatiotemporal features of atmospheric CO? concentrations over Asia, this study conducted a comparation of the spatiotemporal patterns of atmospheric CO? concentrations across Asia from 2003 to 2020 by using three datasets constructed via distinct methods, namely the High-Res, CT2022, and CAMS. The results indicated that there were significant discrepancies in the characterization results of the three datasets. In terms of spatial distribution, all three datasets identified eastern Asia and other regions as high-concentration areas and the Tibetan Plateau and adjacent regions as low-concentration areas, while they exhibited distinct characteristics in terms of mean concentration, seasonal variation regularity and interannual growth rate, with considerable uncertainties in the seasonal growth rates. The discrepancies among the datasets presented obvious spatiotemporal differentiation features: South Asia and Southeast Asia were the core regions with the most prominent discrepancies, and 2014 emerged as a critical node for the variations in discrepancies. The differences between CT2022 and High-Res were significantly affected by seasons, whereas CAMS consistently yielded higher CO? concentration values than High-Res, with more pronounced discrepancies in summer and autumn. This study initially clarified the characterization features of each dataset, which provides a scientific basis for the selection and calibration of relevant datasets, and also offers reliable data support for the assessment of carbon sources and sinks, simulation of carbon cycles, and formulation of low-carbon emission reduction policies in the Asian region.