One hundred and thirty-eight oil samples have been analyzed by visible and near-infrared transflectance spectroscopy. These comprised 46 pure extra virgin olive oils and the same oils adulterated with 1% (w/w) and 5% (w/w) sunflower oil. A number of multivariate mathematical approaches were investigated to detect and quantify the sunflower oil adulterant. These included hierarchical cluster analysis, soft independent modeling of class analogy (SIMCA method), and partial least squares regression (PLS). A number of wavelength ranges and data pretreatments were explored. The accuracy of these mathematical models was compared, and the most successful models were identified. Complete classification accuracy was achieved using 1st derivative spectral data in the 400-2498 nm range. Prediction of adulterant content was possible with a standard error equal to 0.8% using 1st derivative data between 1100 and 2498 nm. Spectral features and chemical literature were studied to isolate the structural basis for these models.