Recently, the authors proposed a new approach for predicting a wide range of physical and thermodynamic properties, based on similarity between molecular structures. This presentation includes the rigorous QS2PR (Quantitative Structure - Structure - Property Relationship) method and its short-cut version for members of homologous series, and the Targeted QSPR (TQSPR) method. Some latest, yet unpublished results, on their application for prediction of interaction parameters between pure components in thermodynamic models, needed for phase equilibria calculations, are also presented. All methods advocated are based on the assumption that similarity between molecular structures can be expressed as a relation between vectors of molecular descriptors, describing these structures. Their main advantage is that the properties of a target compound are predicted from the properties of compounds, the structures of which are proven to be similar to that of the target. They also provide for estimation through consistency analysis of the error with which the properties are predicted.