The final color of coffee roast is one of the most important parameters to control the roasting degrees in industries. Commercial roasting degrees are commonly known as light, medium and dark. Another specific descriptor is Agtron value. Our goal in this research was to evaluate the capacity to predict Agtron values and classification of roasting degrees in whole and ground coffee using colorimetric parameters. Thirty raw coffee samples (being ten samples in each roasting degrees) were roasted with Agtron values between #75 and #65 = light; #65 and #55 = medium; #55 and #45 = dark. The Agtron value was evaluated by spectrophotometer M-basic II (Agtron Inc.) and color was measured in the L*, a*, b* system using colorimeter CM-5 (Konica Minolta Sensing Inc.). Principal Component Analysis (PCA) was applied to discriminate the samples. Partial Least Square (PLS) regression was used for prediction and Partial Least Square-Discriminant Analysis (PLS-DA) for classification. All models were built using the Chemoface program. PCA was able to discriminate the coffees based on the three roasting degrees with 97.07% explained variance for whole coffee and 98.11% explained variance for ground coffee. PLS models obtained good results for whole coffee in validation (R²cv = 0.98; R²test = 0.93), and also for the ground coffee (R²cv = 0.99; R²test = 0.99). All models had a great prediction performance. PLS-DA models were evaluated using the percentages of correct classification (accuracy) in external validation, it was for whole coffee light = 75%; medium = 66.67%; dark = 100% and for ground coffee light = 100%; medium = 66.67%; dark = 100%. This last approach was better to classify the coffees in all roasting degrees. The color parameters with chemometric analysis can be successfully used to classify the roasting degrees and predict the Agtron value in whole and ground coffees.