A new method for the discrimination of adulterated milk based on two-dimensional(2D) correlation infrared spectroscopy and least square support vector machines (LS-SVM) was proposed. 48 pure milk samples were collected and 16 urea-tainted milk (0.01~0.3 g/L), 16 melamine-tainted milk (0.01~0.3 g/L), 16 tetracycline-tainted milk (0.01~0.3 g/L) were prepared. Based on the characteristics of 2D correlation infrared spectra of pure milk and adulterated milk, 6 apparent statistic parameters of all samples were extracted and calculated. These 6 parameters were used as input for LS-SVM to build discriminant model of adulterated milk and pure milk. The recognition rate of unknown samples was 90.6%. The results reveal that parameterization of 2D correlation spectra in combination with LS-SVM method has a feasible potential to discrimination adulterated milk and pure milk.