This paper presents a discrete modeling approach that is based on clusters of molecules. The aim of the approach is to provide the excess Gibbs function (gE) resp. activity coefficients of condensed phase mixtures, in particular, liquids. The modeling approach uses the probabilities of the discrete states of molecular clusters as its main variables. The clusters consist of four molecules. The compounds are modeled as dice-like molecules, giving them the option of having one to six different energetic interaction sites. A molecular sampling algorithm links the abstract dice-like representations to real molecules. The model results are compared to experimental data for mixtures of butanal + n-heptane, acetone + n-heptane, and acetone + methanol. The comparison shows that the deviations of this approach are similar to those of the well-known UNIFAC model, which indicates the approach as a promising alternative for the development of gE models.