Abstract: In order to explore a approach to measure vitamin C content of orange, based on wavelet transform by different decomposing levels, the near-infrared spectroscopy signals of 100 intact orange samples were de-noised and some PLS-CV(partial least squared-cross validation) operations were proposed for the prediction of orange VC(Vitamin C) content with the reconstructed spectra after de-noised. The results show that the PLS-CV results were not the same when the wavelet decomposing level was different. PLS-CV result was the best at a wavelet decomposing level of 4. Its R was 0.9574, and its RMSECV was 3.9 mg/(100 g). Therefore, it is concluded that the FT-NIR model treated by wavelet de-noised is feasible to detect VC content of orange rapidly and nondestructively.