The use of LNG (liquefied natural gas) as fuel brings up issues regarding safety and acceptable risk. The potential hazards associated with an accidental LNG spill should be evaluated, and a useful tool in LNG safety assessment is computational fluid dynamics (CFD) simulation. In this paper, the ADREA-HF code has been applied to simulate LNG dispersion in open-obstructed environment based on Falcon Series Experiments. During these experiments LNG was released and dispersed over water surface. The spill area is confined with a billboard upwind of the water pond. FA1 trial was chosen to be simulated, because its release and weather conditions (high total spill volume and release rate, low wind speed) allow the gravitational force to influence the cold, dense vapor cloud and can be considered as a benchmark for LNG dispersion in fenced area. The source was modeled with two different approaches: as vapor pool and as two phase jet and the predicted methane concentration at sensors' location was compared with the experimental one. It is verified that the source model affect to a great extent the LNG dispersion and the best case was the one modeling the source as two phase jet. However, the numerical results in the case of two phase jet source underestimate the methane concentration for most of the sensors. Finally, the paper discusses the effect of neglecting the -9.3 degrees experimental wind direction, which leads to the symmetry assumption with respect to wind and therefore less computational costs. It was found that this effect is small in case of a jet source but large in the case of a pool source. (C) 2012 Elsevier Ltd. All rights reserved.