db5小波。By using db1～db10 (10 in total) wavelet transform with the decomposing level of 4, the near-infrared spectra (NIRS) signals obtained from 120 umbilical orange samples were denoised, the PLS models of vitamin C (VC) content were established by each wavelet base. Through comparing the prediction accuracy of PLS models, it shows that all db wavelet transform can eliminate the noise in VC NIRS effectively, the prediction accuracy of PLS models are higher than models not being treated. When the wavelet was db5, the best prediction effect was obtained, with the correlation coefficient R between the prediction and true values being 0.9427 and the expected variance RMSECV being as low as 2.02mg/(100g). In conclusion, the db5 is the best wavelet for NIRS denoising of VC content in umbilical orange.