Based on the method of kernet Orthogonal Projection to Latent Structure Discriminant Analysis, discrimination models for adulterated milk were established in the present paper. Forty adulterated milk samples with melamine（0.01～3 g·L-1）and 40 adulterated milk samples with urea （1～20 g·L-1）were prepared, respectively. Then the near-infrared absorption spectra of all samples were measured. The spectra in the range of 4 200～4 800 cm-1 were selected to construct the KOPLS-DA models for milk adulterated with melamine, milk adulterated with urea and milk adulterated with both melamine and urea. The results showed that, compared with PLS-DA and OPLS-DA models, KOPLS-DA model had better discriminant ability for the adulterated milk, and its classification accuracy rate (CAR) for milk adulterated with melamine, milk adulterated with urea and milk adulterated with both melamine and urea were 95%, 100% and 97.5%, respectively.