Abstract: Carbon monoxide (CO) is an essential component of air, and it arises from anthropogenic sources and natural sources. These different sources of CO have significant differences in carbon and oxygen isotopic values, therefore the carbon and oxygen isotopic composition of CO is important for the sources and sink of CO in environmental research. In addition, the isotopic composition of CO in coal beds can help to explore the origin and deposit of coal, the self-heating and spontaneous combustion of coal, which benefits for coal mine safety management. At present, CO isotopic compositions are mainly analyzed by optical spectroscopy or gas chromatography-isotope ratio mass spectrometer (GC-IRMS). The optical spectroscopy method is mainly used in the studies on theoretical CO isotope effects, but its analytical precision is low. GC-IRMS for CO isotopic composition analysis usually requires manual injection, resulting in low efficiency, and it is easily influenced by operation. Meanwhile, both N₂ and CO₂ will produce interfering ions and affect the accuracy of CO isotope analysis. In this study, a device of automatic CO gas separation, purification, and injection was developed and combined with IRMS for determination of isotope composition of CO. The impurities, CO₂ and water molecules in the mixed sample gas were discharged by Nafion tube and a liquid nitrogen cold trap. After nitrogen and CO were separated by Molesive5A chromatographic column, nitrogen was eliminated by valve switching technique, and therefore only the purified CO entered the IRMS for analyzing. By calling the reserved interface of IRMS, the automatic lifting of the cold trap and switching of the 8-pass valve and 4-pass valve was realized. The CO gas concentration and the total peak area of ions at m/z 28, 29 and 30 show a good linear relationship with the correlation coefficient (R²) of 0.999 when the CO gas concentration ranges from 300 to 20 000 μL/L. The standard deviations of δ¹³C and δ¹⁸O are within 0.4‰ when the CO concentration is between 1 000 and 20 000 μL/L. In the repeating analysis of a long term (about 6 months), the results of δ¹³C and δ¹⁸O of 4 000 μL/L CO standards are stable and the standard deviations are 0.26‰ and 0.47‰ (n=20), respectively. The standard deviations of δ¹³C and δ¹⁸O of the CO in natural samples from Coal Bed Methane are less than 0.5‰. Based on the above test results, the conclusions can be obtained that the accuracy and stability of the automatic injection equipment coupled to IRMS meet the requirements for the determination of carbon and oxygen isotopes in CO. However, when CO gas concentration is lower than 1000 μL/L, the carbon and oxygen isotope analysis are greatly biased by IRMS. Further protocol is needed to improve the accuracy and precision of isotopic analysis for low-concentration samples.