The desorption isotherms and thermodynamic properties of coffee from different processing stages were obtained during the drying process of this product. The isotherms were determined by a static gravimetric method for various temperature and humidity conditions. Equilibrium moisture content (Me) data were correlated by several mathematical models and an artificial neural network (ANN) model. The Me for coffee fruits, pulped coffee and green coffee increased with an increase in the water activity (aw) at any particular temperature. At a constant aw, coffee fruits samples had higher Me than the remaining coffee samples. Based on statistical parameters, the ANN model, modified Henderson and GAB models were adequate to describe the sorption characteristics of the samples. Isosteric heat of sorption was evaluated by means of the Clausius–Clapeyron equation. It decreased with increasing moisture content. The coffee fruits had higher isosteric heat of sorption than that pulped coffee and green coffee.