Abstract: Kokotireke grapes, a geographical indication agricultural product in Xinjiang, exhibit the challenges in uneven quality grades and geographical indication adulteration in the market. The visible/near-infrared (Vis/NIR) spectroscopy technology was innovatively employed to perform the quality prediction of Kokotireke grapes and geographical indication identification in this study. Kokotireke grape samples with geographical indication and non-geographical indication were collected, and the spectral data, soluble solid content (SSC) and titratable acidity (TA) were measured. Subsequently, different spectral preprocessing techniques combined with the partial least squares regression (PLSR) to construct quality prediction models and partial least squares discriminant analysis (PLS-DA) geographical indication identification models. The results showed that different preprocessing methods exhibited a significant effect on the model performance, and the SG-LBC-1stD had the best preprocessing effectiveness. Specifically, the determination coefficients (R²) of PLSR model for SSC and TA prediction reached 0.9361 and 0.9478, respectively, and the root mean square errors (RMSE) were 0.3362 and 0.0368, respectively. Additionally, the PLS-DA model can effectively distinguish the products between geographical indication and non-geographical indication, achieving the identification accuracy of 0.928, R²_X of 0.828, R²_Y of 0.621, Q² of 0.553, and AUC of 0.980. Overall, this research provides a rapid, online, efficient and non-destructive quality prediction and geographical indication discrimination method for the Kokotireke grape industry, which is of great significance for ensuring product quality and avoiding the geographical indication adulteration and also offers the reference for related research in other fruits.