Near-infrared (NIR) spectroscopy has gained wide spread acceptance in recent years as a powerful diagnostic tool, particularly for the concentrations of Components of Organic Materials. Unfortunately, most systems in practice are not perfectly linear; they show non-linear behavior of different types. Any model's capacity of information is not infinity, and it is founded that the calibration model is of a strong selectivity caused by non-linear, i.e. the model will provide satisfactory predict results for the samples in the appropriate ranges of the measurements, but make poor ones for the samples in the regions, especially for the extremes of the measurements. Reality has thus created a need for methods that can handle such non-linearities. A new technique of Multi-Region Model (MRM) instead of unique model is presented in the work. To validate the calibration, 198 milk samples were employed in this study, the comparison with the MRM method was based on the root mean square error of prediction (RMSEP) and Correlation coefficient (R2). The study result shows that the MRM accuracy for individual component's prediction is reliable. The predicted results of MRM exhibit values of R2 of 98.63% and 95.07%, and RMSEP of 0.116% and 0.101% for fat and protein, respectively.