The study examined the effect of the choice of reference method on the functionality and reliability of calibrations in near-IR (NIR) spectroscopy intended for measuring the fat content in raw cow's milk. The fat content in the milk samples was evaluated using methods according to either Röse-Gottlieb or Gerber. The same samples were then subjected to analysis on an Antaris FT-NIR spectrometer. Using a partial least-squares algorithm, calibration models were created for both methods from the values measured. The calibration models show very good values of standard error of calibration: 0.133 for the Gerber method and 0.095 for the Röse-Gottlieb method. These calibrations were subsequently used to analyze 30 new samples of cow's milk of undefined fat content, and the differences in the values were evaluated using statistical paired t-test to a median value at a probability level of α = 0.05. No statistically significant differences were found. It was revealed that the reference method used for calibrating the device evaluating the fat content in raw cow's milk has no effect on the functionality and reliability of the calibration model.