Running vapor-liquid equilibrium (VLE) data correlation with interaction parameters independent of temperature is a common practice that may lead to incongruous situations, which may remain unnoticed. A non-negligible number of VLE data sets for binary systems appear very poorly correlated in literature by using any of the existing models with constant parameters and without any apparent explanation. This paper illustrates the unavoidable necessity of considering the temperature dependence for the parameters of the models used to formulate the activity coefficients of the liquid mixtures (such as NRTL, UNIQUAC, Wilson, among others) in order to adequately represent certain experimental data. For these systems, such an approach is mandatory and not a choice. Moreover, there are certain cases where even considering such temperature dependence of the parameters of the liquid phase is not enough and other alternatives must be used in order to enable a coherent correlation of the VLE data. In addition, a strategy has been suggested that allows knowing in advance when the parameters should be considered temperature dependent. The attractive feature of this proposal is that the answer to the question of whether a T-dependence is necessary is obtained just from the experimental VLE data, without using any particular model for the Gibbs energy of mixing for the liquid phase. New results of VLE data correlation showing markedly improved fittings are presented for some of the systems poorly correlated in literature, showing the potential of the procedure. (C) 2018 Elsevier B.V. All rights reserved.