Near-Infrared hyperspectral imaging (HSI-NIR) is a useful technique for pharmaceutical research and industry alike. It can provide important surface information such as the polymorphs quantification and its distribution over the tablet. Several chemometric tools are applied for this purpose, with MCR-ALS and PLS regression being the most common approaches. In this work, a detailed comparison between these two approaches is performed. Beyond a “simple” regression comparison, a comparison of the score images (local quantification) was also evaluated. The system under study is tablets with ternary mixtures of Mebendazol (MBZ) polymorphs, microcrystalline cellulose and magnesium stearate. PLS models, in general, gave lower RMSEP (below 1.7% w/w for the three MBZ polymorphs) than the corresponding MCR-ALS predictions. Analyzing the distributions of the scores in the images of each sample shows clear differences between the PLS and MCR-ALS models. The MCR-ALS gave more chemical meaningful distribution maps for all polymorphs, even though the PLS accurately predicts the average concentration across the image. The problem is that the PLS models used the main spectral regions to quantify each MBZ polymorph, but at the same time undermines the minor spectroscopic changes caused by the different polymorphs. Although this may seem as a minor deviation from the truth, the results clearly show that this deviation is detrimental for the analysis of the spatial distribution of the analytes. These results indicate that the optimal multivariate model for multivariate images depend on the goal of the analysis: global quantification or a distribution analysis.