The performance of several semi-empirical expressions for correlating the temperature, pressure and composition dependence of the thermal conductivity (λ) of pure organic liquids and mixtures was investigated. The temperature and pressure dependence is adequately represented by Chisholm approximants of order (1, 1) or (2, 1) with five and eight adjustable constants, respectively. The fully predictive Vredeveld equation uses mass fractions as the composition variable. It significantly underestimated λ values for the R32 + R125 + R134a ternary refrigerant system. Binary predictive models with one or two adjustable parameters include the quadratic Scheffé polynomial and its corresponding Padé approximant, the cubic "Margules" model and the theoretical Wassiljewa equation. It was found that the Padé (2, 2) approximant and the Wassiljewa equation satisfactorily correlated the extensive ternary mixture data published by Rowley and coworkers. Best results were obtained when the mole fraction was used as a composition variable. The predictive capability of the models was checked using the R32 + R125 + R134a ternary refrigerant system. Combining rules were used for cross parameters such that the temperature and pressure dependence was incorporated via the pure fluid properties. Model parameters were fixed using binary data alone. In this case, the quadratic Scheffé, Padé (2, 2), and Wassiljewa (with temperature-and pressure-independent parameters) all provided satisfactory predictions for ternary mixtures.