Abstract: Multi-cultivar modeling using visible-near-infrared spectroscopy (VIS-NIR) was explored for the rapid non-destructive detection of the internal quality of kiwifruit during storage. In this study, 'Hayward' 'Jin Tao' and 'Xu Xiang' kiwifruit were used as the experimental subjects to assess changes in hardness, soluble solids, titratable acid, and flesh color under different storage times. Spectral data were collected at wavelengths ranging from 592~1102 nm. After the use of different preprocessing algorithms, such as first-order derivatives (FD), standard normal variate (SNV), second-order derivatives, convolutional smoothing, and FD+SNV, the data were combined with competitive adaptive reweighted sampling (CARS) for feature wavelength selection. A quality prediction model based on partial least squares (PLS) and multiple linear regression (MLR) was developed for kiwifruit physicochemical indices. The results showed that the FD and SNV-preprocessed models had the highest prediction accuracies. The relative prediction deviations (RPDs) of SSC in the single-cultivar model all exceeded 2.3. For hardness, while the RPD of Xu Xiang was 1.8, those of other cultivars exceeded 2.3. CARS was used to extract the 600~700, 930~990, and 1000~1100 nm bands with high correlation. The PLS model predicted relatively better performance than the MLR model for each indicator. The establishment of a generalized model for mixed cultivars resulted in significantly enhanced predictive performance with FD+SNV combined with preprocessing, yielding RPDs of 2.280, 2.183, and 3.425 for the SSC, TA, and a^* models, respectively, which demonstrated superior accuracy compared to the single-cultivar model. These findings indicate that VIS-NIR spectroscopy can be used for the quantitative detection of internal quality of kiwifruit during storage, providing a basis and reference for the application of non-destructive testing technology in kiwifruit.