In this paper, Computational Fluid Dynamics (CFD) approach was used to study patterns of Liquefied Natural Gas (LNG) vapor dispersion under different atmospheric stabilities. First of all, the Burro 8 field test was used to validate prediction results generated by Fluidyn-PANACHE, a CFD modeling software. The test data confirmed that PANACHE is able to describe the dense gas behavior of LNG vapor dispersion. The CFD predictions regarding lower flammable limit (LFL) distances and downwind maximum concentration were in approximate agreement with the test data. After this, the dispersion of LNG vapor, which results from its accidental release from the top of large-scale storage tank under different atmospheric stabilities, were simulated. The results showed that the influence of atmospheric stability on the vertical and lateral distributions of vapor cloud, the downwind near-ground gas concentrations and the LFL distance is significant. The vertical spread of cloud under unstable condition is slightly larger than that under neutral and stable conditions. The dilution of the LNG cloud is suppressed under stable condition due to weaker turbulent mixing. Also, stable atmospheric condition can increase the near-ground gas concentrations, the lateral distribution of the cloud and the LFL distance. The longitudinal distribution of the cloud near the ground under neutral condition is larger than that under unstable and stable condition.