Melamine (2,4,6-triamino-1,3,5-triazine) contamination of food has become an urgent and broadly recognized topic as a result of several food safety scares in the past five years. Hyperspectral imaging techniques that combine the advantages of spectroscopy and imaging have been widely applied for a variety of food quality and safety evaluations. In this study, near-infrared (NIR) hyperspectral imaging technique was investigated to detect low levels (<= 1.0%) of melamine particles in milk powders. Following image preprocessing (normalization and background removal), the spectrum of each pixel in the sample images was compared to the pure melamine spectrum by spectral similarity measures including spectral angle measure (SAM), spectral correlation measure (SCM), and Euclidian distance measure (EDM). The three similarity analysis methods provided comparable results for melamine particle detection where imaging allowed visualization of the distribution of melamine particles within images of milk powder mixture samples that were prepared with various melamine concentrations. The classification results were verified by spectral feature comparison between separated mean spectra of melamine pixels and milk powder pixels. The study demonstrated that a combination of NIR hyperspectral imaging technique and spectral similarity analyses was an effective method for melamine adulteration discrimination in milk powders. The method described in this study can also be applied to other chemicals or multi-chemicals adulterant detection in milk powders. Published by Elsevier Ltd.