Abstract: Precise measurement of carbon isotopes is crucial for investigating the sources, cycling, and emissions of carbon in water bodies. Common methods for carbon isotope measurement primarily include mass spectrometry and optical spectroscopy. Although isotope ratio mass spectrometry (IRMS) offers high precision, it has limitations such as expensive equipment and complex operational procedures. Cavity ring-down spectroscopy (CRDS) has been widely applied in carbon isotope analysis due to its high sensitivity and rapid real-time analysis capabilities, which shows high measurement precision approaching that of IRMS. However, factors affecting the CRDS measurement precision have not been sufficiently investigated. Particularly, a knowledge gap remains in the quantitative characterization of matrix interference effects, which limits its accuracy in the practical analysis of complex environmental samples. To address these issues, this study established a direct gas sample determination procedure by modifying the CRDS inlet system, and systematically evaluated the effects of CO₂ concentration, instrument operation time, and background gas composition on measurement results. Results showed that the CRDS system achieved a measurement precision of ±1.5‰ for CO₂ carbon isotopes. Among the influencing factors, instrument operation time was the most dominant factor affecting measurement precision (F=25.62, P＜0.001), with δ¹³C values showing systematic negative drift as operation time increased. The underlying cause was potentially linked to stress changes in optical cavity mirrors and minute cavity displacement. O₂ in the background gas exhibited significant matrix interference effects (F=7.46, P=0.008), while CO₂ concentration had minimal impact on the measurement results. To ensure measurement accuracy, testing within 2−4 hours of instrument operation is recommended, with O₂ content in samples minimized as much as possible. The CO₂ carbon isotope values of groundwater gases from the Datong Basin determined by this technique ranged from −9.25‰ to −32.57‰, which were generally consistent with external validation results, demonstrating the stability and effectiveness of this method.