Coals are known to be often associated with gas and condensate accumulations. Yet experimental evidence suggests that they generate as much heavy compounds as conventional type II kerogen. The concepts brought forward to date to explain this contradiction are rather unspecific and not quantifiable. This paper investigates the role of adsorption in nanopores of vitrinite and suggests quantifiable processes of retention. We combined computer process models of nanopore adsorption and absorption with a generic ten-component generation and cracking scheme in order to model generation, retention and cracking of petroleum component groups in vitrinite. A vitrinite adsorption model based on Dubinin-Radushkevitch theory expels a highly aromatic gas-condensate. An absorption model based on Polymer Solution theory fails to eliminate heavy compounds sufficiently. A combination of adsorption and absorption processes, however, explains why coal expels gas, even if its chemical model suggests generation of large amounts of C15+ hydrocarbons. The inherent simplifications of the model seem not to influence the general direction of fractionation but its magnitude. The results obtained to date suggest that nanopore distribution and cross-link density in macerals may influence the composition of expelled petroleum in a decisive manner. Proper quantitative predictions need more accurate knowledge of multi-component swelling ratios and adsorption systems, vitrinite pore wall polarity at different maturities, and more detailed vitrinite pore size distributions. (c) 2004 Elsevier B.V. All rights reserved.