A study of the Near Infrared Spectroscopy of rice wine is studied, with the purpose to discuss the impact of spectral pretreatment methods on wine alcohol rapid quantitative model, and to compare and analyze the effects of Partial Least Squares quantitative model of spectroscopy after the five traditional methods which include smoothing, first derivative, second derivative, multiplicative scatter correction and standard normal transform and the three new methods which contain wavelet transform, Fourier transform and orthogonal signal correction. The results indicates that the cross-validated R=0.998 and RMSECV=0.135 7 after the orthogonal signal correction is superior to the other pretreatment methods. Therefore, pretreatment with the noise reduction performance is superior to pretreatment with the sharpening performance which illustrate that the random noise of the spectrum is a major influence factor on model accuracy. The results provide an effective pretreatment method for establishing a stable near infrared spectroscopy fast quantitative model for alcohol content of rice wine, and it also offers a reference for rice wine quality online fast detection.