[en] Adsorption of Volatile Organic Compounds (VOCs) by Granular Activated Carbon (GAC) is a conventional process for recovery of solvents from gas effluents. Such a treatment is achieved in cyclic batch adsorption systems, in which the adsorbent bed is alternately saturated and regenerated. However, the exothermal nature of adsorption causes a reduction in the dynamic adsorption capacity of the bed, and thus a decrease in the overall efficiency of the recovery process. Furthermore, the local temperature rises may become hazardous at high levels of pollutant concentrations, bringing about safety risk with bed ignition. This study points out the properties of VOCs and GACs that play a major role on the heat flux released during adsorption. For that purpose, a statistical analysis was conducted on a database representative of the changes in the molar integral adsorption enthalpies measured on 8 different commercials GACs, for a wide variety of VOCs. The adsorption enthalpies were determined by TG-DSC (thermal gravimetry analysis coupled to differential scanning calorimetry), by flowing helium loaded with 50 g.m³ of solvent through the adsorbent sample. Afterwards, two statistical models were applied on the database so generated and were compared in terms of prediction skill: Multi-Linear Regression (MLR) and Neural Network (NN). From the MLR approach, the polarizability, the heat of vaporization, the ionization potential, and the surface tension of the VOC were discriminated among other molecular properties, whilst the average micropore radius was shown to be the most influent adsorbent characteristic on the potential energy of the interactions with an organic molecule. The major influence of these properties was explained in the light of the mechanisms that are involved during progressive coverage of the adsorbent. Physisorption, i.e. dispersion forces between the adsorbate and the pore walls, and capillary condensation occurring in meso-pores were put forward. The MLR approach enabled to compute integral adsorption enthalpies with a precision around 10%. Applying a NN approach to compute the adsorption enthalpies did not improve the predictive ability. (authors)