Liquefied natural gas (LNG) rollover refers to the sudden mixing of stratified LNG layers, which can cause the generation of significant amounts of boil-off gas. Such events are a significant safety concern in LNG storage but there are no reliable models for its description at industrial scales available in the open literature. In this article, the data and models for LNG rollover existing in the open literature are reviewed and a new framework for quantitatively analyzing the limited available data is presented. We extended the definition of the hydrostatic stability ratio for binary mixtures to allow its estimation for multicomponent mixtures, either from the reported LNG layer compositions or measurements of the LNG layer densities. By analyzing the graphical data of Bates and Morrison (Int J. Heat Mass Transfer. 1997;40:8) the critical value of the stability ratio, R-c, separating the diffusive phase of LNG rollover from the penetrative convection phase was estimated to be 3.8 +/- 0.5. This is significantly larger than the critical ratio of 2 reported for saline solutions and is also larger than the initial stability ratio of 1.7 estimated from the best documented LNG rollover incident at La Spezia in 1971. Lumped-parameter models for LNG rollover reported in the literature have successfully described the La Spezia incident by using the Reynolds analogy to estimate mass transfer rates from heat transfer correlations. However, these same models are unsuccessful when applied to other reported LNG rollover incidents, with the predicted rollover time being too short because the mass-transfer coefficient is overestimated. The results presented here suggest that these limitations could be overcome by using a smaller mass-transfer coefficient (estimated, e.g., from the Chilton-Colburn analogy) and by tracking the multicomponent system's stability ratio until the critical value is reached whereupon the mass transfer regime changes. (c) 2013 American Institute of Chemical Engineers