Abstract: We have applied partial least squares (PLS) to establish a calibration model, the precision of which is verified by cross-validation. Via adopting optimum wavelength to calibrate, a satisfactory model has been established. In this model, the correlation coefficient in methanol determination is 0.999 91, the root mean square error of cross-validation (RMSECV) is 0.431; in ethanol de termination, the correlation coefficient is up to 0.999　98, and the RMSECV is 0.193. Compared with that of gas chromatography, the relative error of the sample determination results is not more than 3.505%.

Key words: near infrared spectrum, partial least squares, methanol, ethanol