【Objective】 The soil carbon-to-nitrogen ratio (C/N) reflects not only soil quality but also the nutrient balance of soil carbon and nitrogen elements. Thus, rapid and accurate determination of this ratio and the grade is crucial for guiding real-time scientific fertilization and improvement of soil quality. 【Method】 This study used visible-near infrared (VNIR) and mid-infrared (MIR) spectroscopic data, along with total organic carbon (TOC), total nitrogen (TN), and C/N data from 501 typical tobacco-corn rotation farmland topsoil samples(0～20 cm) in Guizhou Province for characterization. After processing the spectra with Savitzky-Golay (SG) smoothing and standard normalization, three modeling methods were applied: partial least squares regression (PLSR), random forest (RF), and Cubist. Models for predicting soil C/N were constructed using both direct prediction of C/N and indirect prediction (first predicting TOC and TN, then calculating C/N), and the precision of C/N value and grade predictions was analyzed. 【Result】 The results revealed that: (1) For C/N value prediction, the optimal prediction strategy was direct prediction using MIR-PLSR, which had a prediction precision(relative standard error, RPD) of 1.20; (2) C/N grade could be accurately predicted, with the optimal strategy being direct prediction using the MIR-PLSR model, achieving a grade determination accuracy of 0.71; (3) The main reasons for the low prediction accuracy of C/N values are twofold. First, the uniform stringent fertilization measures in the tobacco fields have reduced the spatial variation in the carbon and nitrogen content of the plow layer soil, thereby also reducing the spatial variation of C/N(the coefficient of variation is 17.15%, indicating moderate variation). Second, the correlation between C/N and both VNIR and MIR spectra was relatively low. 【Conclusion】 Therefore, the MIR-PLSR model can be used for direct prediction of C/N grades.