The test particle method has been used in conjunction with molecular dynamics simulations to evaluate the solubility of noble gases in silicate melts of various composition. At low pressure the calculated solubility constants (the inverse of the Henry's constant) are in excellent agreement with data of the literature. In particular it is found that the solubility constant (i) decreases when the size of the noble gas increases, (ii) decreases from silica-rich to silica-poor composition of the melt, and (iii) is positively correlated with the temperature. Moreover it is shown that the solubility is governed primarily by the entropic cost of cavity formation for inserting the noble gas into the melt and secondarily by its solvation energy. Interestingly, the behaviour of these two contributions differ from each other as the entropic cost of cavity formation increases strongly with the size of the solute atom to insert whereas large atoms are better solvated than small ones. Considerations of thermodynamics show that the weight fraction of a noble gas in a silicate melt coexisting with its parent fluid at T and P is equal to n(g)gamma(m)/n(m)gamma(g), where n(g) and n(m) are the densities of the two coexisting phases (gas and melt, respectively) and where the solubility parameters gamma(m) and gamma(g) express the probability of inserting the noble gas atom in the melt and in the parent fluid, respectively. The gamma(m) and gamma(g) decrease drastically when the pressure is increased and the noble gas solubility at high pressure is the result of a balance between these two quantities. Here again, the pressure behaviour of gamma(m) and gamma(g) is dominated by the pressure dependence of the entropic cost of cavity formation, the energetic contribution being of minor importance but not negligible at high pressures. With all melt composition investigated here (silica, rhyolite, MORB and olivine), the calculated solubility curves exhibit the same qualitative behaviour with pressure; a steep rise culminating in a broad maximum followed by a gradual decrease of the solubility at higher pressure. At variance with LHDAC experiments (Chamorro-Perez et al., 1996, 1998; Bouhifd et al., 2006, 2008) where a Ar solubility drop is observed at about 50 kbar in silica and molten olivine and in the pressure range similar to 100-160 kbar with other melt composition, we do not find such a sudden change of the solubility. (C) 2011 Elsevier Ltd. All rights reserved.