A new method for discrimination of oil types by means of near infrared spectroscopy (NIRS) was developed. First, the characteristic spectrums of oil were got through principal component analysis (PCA). The result of the analysis suggests that the reliabilities of first 8 principal components are more than 95.38%. The 2-dimensional plot was drawn with first and second principal components, which indicates that it is a good clustering analysis for classification of oil varieties. Several variables compressed by PCA were used as inputs of multiple discriminant analysis (MDA).150 samples from three varieties were selected randomly, then they were used to build discrimination model. 30 unknown samples were predicted by this model, and the recognition rate is 100%. This model is reliable and practicable. It could offer a new approach to the fast discrimination of oil types.