A very specific descriptor used by the coffee industry to evaluate the roasting degree is the Agtron value. Our goal in this study was to develop an alternative methodology for prediction of the Agtron value of whole and ground coffee using digital image processing and artificial neural network (ANN), and also build a software capable of classifying the roasting degree. Seventy raw coffee samples were roasted and organized into seven classes (ten samples in each class), with Agtron values between #95 (very light) and #85 (light); #85 and #75 (moderately light); #75 and #65 (medium light); #65 and #55 (medium); #55 and #45 (moderately dark); #45 and #35 (dark); #35 and #25 (very dark). The Agtron value was analyzed by spectrophotometer M-basic II (Agtron Inc.). The acquisition of image data was recorded using a computer vision model UVDI-254 (Major Science Inc.) with a digital camera model Power Shot G12 (Canon Inc.). The color histograms with the values means of the RGB variables were obtained using the Fiji ImageJ® software. The ANN structure was built using SAS Statistical Discovery JMP® (SAS Inc.) with a free license for testing. The software, namely FRR 1.0, was build using the Embarcadero® C ++ Builder Community Edition software version 10.2 (Embarcadero Technologies Inc.) and registered at the National Institute of Industrial Property (INPI) with registration number 512019002447-8. The percentages of determination coefficient (R2) for ground coffee were training = 99.97% and validation = 99.97% and for whole coffee were training = 99.83% and validation = 99.94%. All promising results were suggesting the possibility to use the digital image (RGB color) to evaluate the Agtron value. In addition, for more practical application of these adjusted models was developed the FRR 1.0 software, that was able to predict the Agtron value and classify the roasting degree correctly.