Changes in the Earth's climate caused by global warming are a looming problem that poses serious challenges not only for our generation but for future generations. An accurate determination of CO2 gas plays a critical role in this field of research. The measurement of greenhouse gases is pivotal to understanding the changes in Earth's climate and needs to be carried out with a high degree of accuracy. Precision measurements on a 0.1 ¿mol/mol scale may provide research data for precisely monitoring the continuing changes that the planet is undergoing. The World Meteorological Organization (WMO) has recommended that carbon dioxide concentrations in air can be measured by comparing these with national reference gases using a nondispersive infrared (NDIR) analyzer to standardize international data. The CO2 molecules absorb the distinctive resonant frequencies in IR spectrometers. The NDIR analyzers usually use narrow band path filter to determine 12CO2 in all carbon dioxide molecules, which can possibly ignore the measurement of 13CO2 partially or totally. However, if the carbon isotopic abundances of CO2 samples deviate from those in standard CO2 gas, the NDIR measurement will not be exact. For accurate measurements, producers of reference gas mixtures either must use gas with natural isotopic abundances, or report the isotopic abundances of CO2. In order to document shifts based on isotopic variability, we prepared artificial air as CO2 reference gas mixtures gravimetrically with CO2 having different carbon isotopic signatures to study the resulting isotopic variations. We used different δ13C values of two CO2 source gases, A and B, corresponding to -41.97? and -14.88?, respectively, which were measured using an isotope ratio mass spectrometer. One set of reference gas mixtures (A1 to A5) was prepared from the CO2 source of δ13C = -41.97?, and the other set of reference gas mixtures (B1, B2) was prepared from that of δ13C = -14.88?. The CO2 abundances of the two sets of mixtures were compared by using NDIR. The reproducibility test for the set A showed that the data are consistent within uncertainty (calibration line was obtained by the best secondary polynomial least squares fit). The uncertainty of CO2 concentration in the reference gas mixtures are 0.06 ¿mol/mol with a 95% confidence level. The reproducibility of the NDIR measurement is 0.012 ¿mol/mol (standard deviation). The difference between the set A (A1 to A5) and set B (B1, B2) was found to be 0.17 ¿ 0.01 ¿mol/mol, which is in excellent agreement with the theoretically predicted value of 0.17 ¿mol/mol.