An aqueous two-phase system (ATPS) is formed from the thermodynamic incompatibility between protein-carbohydrate pairs under specific conditions of pH, concentration and ionic strength. These systems are evaluated from phase diagrams obtained from physicochemical characterization of the mixtures and separate phases. Although there are several methods for physicochemical characterization, these methods are hard working, produce residues, consume samples and do not always show the best performance. In this sense, the study of low-cost techniques for the characterization of these systems becomes interesting. The objective of this work was the use of NIR-portable to predict the concentration of proteins and carbohydrates in order to build phase diagrams from multivariate calibration models. Initially, mixtures were produced in different concentrations of sodium caseinate and locust bean gum (LBG), which were frozen and lyophilized. After obtaining the dry mixtures, these samples were analyzed in near infrared region (900-1700 nm) using MicroNiR. From the spectra, multivariate calibration models were built using Partial Least Squares (PLS). The PLS models for protein prediction obtained an R2 of 0.88 with a prediction error of 8.7% (w/w), while for the carbohydrate prediction it was possible to achieve an R2 of 0.96 with an error of 1.4% (w/w). Although the models have achieved good fitting, when they are used to build the phase diagrams there is a deviation in the slope of the tie-lines. This deviation was attributed to the migration of NaCl between phases, which, despite not showing a signal in the infrared region, can have a shielding effect and uneven migration between phases. Therefore, the results demonstrate that it is feasible to obtain phase diagrams using low cost NIR, although the effect of NaCl addition should be further investigated.