Coffee is a very popular beverage, consumed worldwide. In Brazil, it is one of the traditional cultivation and the quality of its beverage has been the subject of several studies. The quality of the beverage depends on the previous operations for processing, such as type of cropping, grain maturation, preparation and drying of the coffee grain. In 2004, the Brazilian Association of Coffee Industry (ABIC) established criteria, from which it is possible to classify the coffee into four categories according to its quality level - Not recommended for delivery, Traditional, Superior and Gourmet. The aim was to build multivariate regression models to predict the overall quality of the coffee beverage using the partial least squares methods (PLS) and principal component regression (PCR) assessment of overall quality. For this, 213 coffee samples were submitted shall sensory analysis in the years 2014 and 2015. Using quantitative descriptive analysis method (ADQ) of coffee, with a selected and trained team composed of at least five judges per session, using unstructured 0-10 cm scale for evaluation of powder fragrance, aroma, defects, acidity, bitterness, flavor, aftertaste, astringency and body of the coffee beverage with final evaluation of the overall quality. Of the 213 samples analyzed, the results 195 for making the prediction model and the results of 18 for external validation were used. The exploratory multivariate analysis (PCA) was applied in the ADQ data and the sum of principal components 1 and 2 explained 98.33% of the variance of the data. They were built two models for the coffee global quality forecast through the PLS and PCR methods. Both models showed similar results, but PLS showed better predictive capacity since those correlation coefficients (rcal and rval) had optimal settings.