The g(E) model LIQUAC and the corresponding group contribution model LIFAC are widely used for the reliable prediction of vapor-liquid equilibria (VLE), osmotic coefficients, and mean ionic activity coefficients for aqueous systems that contain strong electrolytes up to high concentrations. These models consist of a Debye-Huckel term, a Pitzer type virial equation for the middle-range contribution, and the UNIQUAC, respectively, the UNIFAC term. However, with the variable reference state proposed herein, a reliable prediction of liquid-liquid equilibria (LLE) using the LIQUAC or the LIFAC model is not possible. Because the mean ionic activity coefficient in all existing liquid phases must be predicted reliably, an explicit fixed reference state for the ions had to be introduced for the normalization of the middle-range term. The aim of this work is to modify the LIQUAC and LIFAC model for the prediction of ternary LLE using interaction parameters based on binary datasets. In addition, the results for binary properties could be improved by further changes of the original models. Using a comprehensive database, the interaction parameters for the modified LIQUAC and the modified LIFAC model had to be refitted and the obtained results for the calculated properties of binary and ternary mixtures were finally compared to the results of the original models.