Cubic equations of state (CEOS) are widely used for process design calculations and reservoir simulations in the oil and gas industry. However, most CEOS yield poor predictions of liquid densities. To remedy this problem, several volume-translation approaches have been presented in the literature. Currently, most of these approaches yield predictions that appear reasonable only in the saturated or the single-phase region. In this work, a volume-translation method is presented that is applicable to both saturated and single-phase regions. The method contains only one fluid-specific parameter, which has been generalized in terms of readily available molecular properties such as critical compressibility factor, acentric factor and dipole moment. Three specific cases were considered ranging from simple linear correlations to non-linear neural networks for generalizing the fluid-specific parameter. The neural network models require 3-6 molecular properties for the selected fluid (in addition to the usual critical temperature, critical pressure and acentric factor). For model development and generalization, a database of highly accurate data for liquid densities was compiled. The database contains 65 pure fluids involving several classes of chemical compounds that vary widely in terms of their molecular size, shape, chain-length, asymmetry and polarity. The model was developed based on these 65 fluids, and an additional 20 fluids were used for validating the generalized model. Results indicate that the volume-translation method developed in this work is capable of precise representations of saturated liquid density data for the fluids in the database. Specifically, an overall average absolute percentage deviation (%AAD) of 0.6 was obtained for 65 fluids, including more than 12,000 data points. The model was then generalized, and the generalization predicted the same data set with a %AAD of 0.8. Further, the generalized model was validated by predicting saturated liquid densities of 20 compounds not used in model development. For these 20 fluids, the method provided generalized predictions with 1.0%AAD. Further, the volume-translation approach was also extended to predict liquid densities in the single-phase region. Of the fluids tested, the generalized model provided predictions with 1.8%AAD. Thus, this approach was capable of producing accurate predictions for both saturated and compressed liquid densities of diverse classes of molecules. (C) 2012 Elsevier B.V. All rights reserved.