Abstract: Xinhui dried tangerine peel fruit pulp was used as a raw material to establish a rapid and quantitative analytical method for monitoring the total soluble sugar (TSS) and nitrogen (TN) contents during carotenoid fermentation by Rhodotorula mucilaginosa (R. mucilaginosa). Near-infrared spectroscopy (NIR) was integrated with partial least squares regression (PLSR) to construct models for quantifying the TSS and TN in fermentation broth. The coefficient of determination for the prediction set (R²p), root mean square error of prediction (RMSEP), and relative percent deviation (RPD) were used as evaluation metrics, and the effects of different spectral preprocessing methods and feature wavenumbers on model performance were systematically compared. The results were externally validated using independent samples. The optimal models for quantifying the TSS and TN content were obtained using multiplicative scatter correction (MSC) for preprocessing the spectrum and competitive adaptive reweighting sampling (CARS) for selecting the feature wavenumber. The model for quantifying TSS achieved a R²p of 0.9670, a RMSEP of 2.6587, and an external validation RPD of 4.8386, indicating high-precision quantitative analysis. The model for quantifying TN yielded R²p and RMSEP values of 0.9502 and 0.0624, respectively, with an external validation RPD of 2.7423, which indicated suitability for routine quantitative analyses. The t-test results for the TSS and TN models indicated that the predicted and measured values did not significantly differ (P>0.05), indicating high accuracy and stability. These models can be used for rapidly monitoring the TSS and TN content in R. mucilaginosa carotenoid fermentation broth samples.