Vis-NIR spectroscopy is often used as a “black-box” without knowing its working mechanism to predict indicators. It is necessary to understand its working mechanism before it can be used reliably. First, Vis-NIR spectroscopy combined with PLSR analysis and the stepwise selectivity ratio (SWSR) method demonstrated the ability to accurately estimate the freshness of komatsuna and broccoli. The monitored cumulative CO2 production of komatsuna and the cumulative temperature of broccoli under different storage conditions were used as indicators of freshness, and the measured Vis-NIR spectra of komatsuna and broccoli were used as predictors. Based on the informative wavelengths (IWs) selected by the SWSR method, accurate predictive models of the freshness of komatsuna (R2p = 0.678) and broccoli (R2p = 0.753) were constructed through PLSR analysis. Then, NMR metabolomics analysis was used to measure the metabolites in the vegetable samples. Through the correlation analysis between the signals of informative NIR spectra selected at IWs and the signals of NMR spectra, four amino acids were commonly identified as candidate freshness markers of komatsuna and broccoli. These identified amino acids were negatively related to freshness, and protein absorption in NIR spectra. Protein degradation to amino acids owing to senescence during vegetable storage was detected as a change in the NIR spectral absorbance, which was chosen to build the predictive model for freshness estimation.