Visible and near-infrared reflectance spectroscopy (NIRS) was applied in the discrimination of sugar varieties.NIRS is a pollution-free,rapid,quantitative and qualitative analysis method,with the characteristics of high speed,non-destructiveness,high precision and reliable detection data,etc.Four kinds of sugar were gained from the local market and each species was divided into 40 samples.One hundred twenty samples were used as the training set and the remainders (total 40 samples) formed the prediction set.Samples were scanned by a spectroradiometer within a wavelength region of 325-1 075 nm.Three pre-processing methods were applied on the spectra prior to building the PLS regression model.The multivariable analysis using partial least square (PLS) was applied to abstract characteristics of the pattern.Through full cross validation,11 principal components presenting important information of spectra were confirmed.The correlation coefficient (R),residual variance (RV) and standard error of calibration (SEC) were 0.999 916,0.000 985 and 0.014 538 respectively.Then,these 11 principal components were taken as the input of BP neural network.This model was used to predict the varieties of 40 unknown samples.Through training and prediction,the recognition rate of 100% was achieved by BP neural network.This model has come to be reliable and practicable.Thus,it is concluded that PLS analysis combined with BP neural network is an available alternative for pattern recognition based on the spectroscopy technology.