Numerical and physical requirements to simulations of sub-sonic release and dispersion of light gas in an enclosure with one vent are described and discussed. Six validation experiments performed at CEA in a fuel cell-like enclosure of sizes H x W x L = 126 x 93 x 93 cm with one vent, either W x H = 90 x 18 cm (vent A) or 18 x 18 cm (B) or 1 cm in diameter (C), with a vertical upward helium release from a pipe of internal diameter either 5 mm or 20 mm located 21 cm above the floor centre, were used in a parametric study comprising 17 numerical simulations. Three CFD models were applied, i.e. laminar, standard k-epsilon, and dynamic LES Smagorinsky-Lilly, to clarify a range of their applicability and performance. The LES model consistently demonstrated the best performance in reproduction of measured concentrations throughout the whole range of experimental conditions, including laminar, transitional and turbulent releases even with large CFL numbers. The laminar and the standard k-epsilon models were under performing in the reproduction of turbulent and laminar releases respectively, as expected, as well as in simulation of transitional flows. The laminar model demonstrated high sensitivity to the CFL (Courant-Friedrichs-Lewy) number even below the best practices limit of 40. Three different computational domains and grids were used in order to clarify the influence of mesh quality on the capability of simulations to reproduce the experimental data. It is concluded that physically substantiated choice of CFD model, the control of the CFL number (and released gas mass balance where appropriate), and the mesh quality can have a strong effect on the capability of simulations to reproduce experiments and, in general, on the reliability of CFD tools for application in hydrogen safety engineering. Copyright (C) 2014, Hydrogen Energy Publications, LLC. Published by Elsevier Ltd. All rights reserved.