A real-time automated process control tool for coffee roasting is presented to consistently and accurately achieve a targeted roast degree. It is based on the online monitoring of volatile organic compounds (VOC) in the off-gas of a drum roaster by proton transfer reaction time-of-flight mass spectrometry at a high time (1 Hz) and mass resolution (5,500 m/Δm at full width at half-maximum) and high sensitivity (better than parts per billion by volume). Forty-two roasting experiments were performed with the drum roaster being operated either on a low, medium or high hot-air inlet temperature (= energy input) and the coffee (Arabica from Antigua, Guatemala) being roasted to low, medium or dark roast degrees. A principal component analysis (PCA) discriminated, for each one of the three hot-air inlet temperatures, the roast degree with a resolution of better than ±1 Colorette. The 3D space of the three first principal components was defined based on 23 mass spectral profiles of VOCs and their roast degree at the end point of roasting. This provided a very detailed picture of the evolution of the roasting process and allowed establishment of a predictive model that projects the online-monitored VOC profile of the roaster off-gas in real time onto the PCA space defined by the calibration process and, ultimately, to control the coffee roasting process so as to achieve a target roast degree and a consistent roasting.