For establishing the determination model of vitamin C content of Red Globe Grape using visible / near infrared diffuse reflectance spectroscopy, and evaluating its application value, effects on scaling model results of different chemometry modeling methods, different spectra pretreatment methods and different effective wavelength intervals were discussed, and the samples in prediction set were used to verified the accuracy of the optimal model. The results showed that in the whole spectral range, application of modified partial least squares (MPLS) combined with first derivative, 5 points smoothing, weighted multivariate discrete correction (WMSC) could obtain the optimal calibration models, the cross validation error (SECV) was 0.054 3, the calibration determination coefficient (R²_cv )was 0.920 2, the prediction determination coefficient (R²p) was 0.931 8 and the standard error of prediction （SEP）was 0.050 0, the sum of squared residuals (PRESS) was 0.188 0, the relative analysis of the forecast error (RPD) was 3.640 0. Therefore, applying visible / near infrared diffuse reflectance technique to quick and nondestructive detection of vitamin C in Red Globe grape is feasible, the model is stable and with high precision.