Abstract: The noise of the near infrared spectrum of tomato was eliminated respectively in 11 kinds of spectral preprocessing methods of the wavelet denoising (WD), the constant offset (COE) and so on, WD was examined to be the optimal spectrum preprocessing method by comparing the prediction of the partial least squares (PLS) calibration model, And by optimizing several wavelength ranges of WD NIR spectra, the best wavelength ranges of 11 998.9-6 097.8 cm-1 and 4 601.3-4 246.5 cm-1 was obtained, in this spectral region, PLS model with WD spectrum was established, which gave a correlation coefficient (R) between prediction and actual of 0.930, a root mean square error of cross validation (RMSECV) of 0.466%, and a standard error of calibration (RMSEC) of 0.469%, as well as a standard error of prediction (RMSEP) of 0.260%. The results showed that the FT-NIR model treated by WD is feasible to detect total sugar content of tomatoes rapidly and nondestructively.