Flash point is one of the most important variables used to characterize fire and explosion hazard of liquids. This paper predicts the flash point of miscible mixtures by using the flash point prediction model of Liaw and Chiu (J. Hazard. Mater. 137 (2006) 38-46) handling non-ideal behavior through liquid phase activity coefficients evaluated with UNIFAC-type models, which do not need experimentally regressed binary parameters. Validation of this entirely predictive model is conclusive with the experimental data over the entire flammable composition range for twenty four flammable solvents and aqueous-organic binary and ternary mixtures, ideal mixtures as well as Raoult's law negative or positive deviation mixtures. All the binary-mixture types, which are known to date, have been included in the validated samples. It is also noticed that the greater the deviation from Raoult's law, the higher the probability for a mixture to exhibit extreme (minimum or maximum) flash point behavior, provided that the pure compound flash point difference is not too large. Overall, the modified UNIFAC-Dortmund 93 is recommended, due to its good predictive capability and more completed database of binary interaction parameters. Potential application for this approach concerns the classification of flammable liquid mixtures in the implementation of GHS. (C) 2010 Elsevier B.V. All rights reserved.