The objective of this study was to use the partial least squares regression (PLS) for the development of internal calibration model, to predict the hardness of lipid bases based on the solid fat content. The lipid bases were made from the blend of anhydrous milk fat (AMF) and high oleic sunflower oil (HOSO) in the following proportions (AMF:HOSO): 100: 00; 90:10; 80:20; 70:30; 60:40; 50:50 (w/w). The solid fat content was determined using Nuclear Magnetic Resonance Spectrometer (NMR), with sample conditioning at evaluation temperatures. The hardness of the blends was determined at 5 and 15 °C in the Texture Analyzer (TA-XT2i) using a 45 degree acrylic cone; 10 mm distance; speed of 2 mm/s; and time of 5 s. The chemometric calculations of multivariate calibration, PLS, were carried out using the Pirouette software package 3.1. Thirty-six samples were used to construct the model from 6 treatments, 3 processes, and duplicate analysis. Three latent variables obtained by cross-validation were selected for the lipid bases stored at 5°C and 15 °C. At 5 °C, the addition of up to 20% HOSO to AMF showed high leverage, probably due to the higher milk fat content of the samples. However, at 15 °C only the blend 100:00 leveraged the regression model, probably due to most of HOSO triacylglycerols had been melted at this temperature. For both temperatures, the values of studentized residual were within the expected range (-2.5 to +2.5) for all samples. The correlations between the values obtained and predicted by the model were R2 0.9839 and 0.9719 for the temperatures of 5 °C and 15 °C, respectively, indicating good fit and high ability to predict the hardness of lipid bases from the solid fat content determined by Nuclear Magnetic Resonance.