Saturated water solubility values in hydrocarbon liquids span more than four orders of magnitude and are characterized by high relative error. Development of reliable correlations is particularly challenging for ill-defined and asymmetric hydrocarbon mixtures where fluid characterization and pseudo binary phase behavior type are uncertain. In this work, predictive correlations targeting this application, below the critical temperature of water, are explored and two promising models are compared with computation methods available in the literature. The model inputs include easily measurable properties (hydrogen mass fraction, fluid density at ambient conditions) as well as K-Watson, mean molar mass and the average boiling point of the hydrocarbon mixture. These latter three inputs are complex and frequently difficult to define parameters encountered in refinery characterization of ill-defined organic mixtures. Models comprising one or two input variables were fit using a training data set comprising pure hydrocarbons and Athabasca bitumen. The average deviation of the models from these reference experimental data was found to be comparable to the uncertainty of the reference data in most cases. However most of the models exhibited high deviation and bias relative to the test data set comprising ill-defined hydrocarbon mixtures. Two models are shown to be robust numerically, insensitive to the uncertainty of input variables, and applicable irrespective of the pseudo binary phase behavior type with water (Type II, Type IIIa, or Type IIb) up to their upper critical end point (UCEP). At temperatures above the UCEP, a hydrocarbon rich liquid phase only arises for Type IIb pseudo binary mixtures. In this region, above 647 K, correlation extrapolation options beyond the range of experimental data are discussed, and only fluid-specific Henry-like constant models are proposed. (C) 2014 Elsevier Ltd. All rights reserved.