In this study， near-infrared spectroscopy technology combined with chemometric methods was used to monitor the changes of tea saponin mass concentration， protein mass concentration and polysaccharide mass concentration in the process of tea saponin extraction in real time. The results showed that the first-order derivation(1stDer) was the optimal preprocessing method with the highest RC， RP value and the least RMSECV， RMSEP in all pretreatments. The performance of a model was evaluated by the correlation coefficient (RP) and the rootmean square error (RMSEP) in the prediction set. Compared with partial least squares (PLS) and interval partial least squares (iPLS) modeling， the model built under the synergy interval partial least squares (Si-PLS) algorithm had the best robustness. The tea saponin concentration RP=0.9889， RMSEP=1.36； protein concentration RP=0.9859， RMSEP=0.354； polysaccharide concentration RP=0.9919， RMSEP=0.359. Conclusion: Near-infrared spectroscopy technology combined with Si-PLS model can better monitor the extraction process of tea saponin in real time.