To numerically simulate the convective melting of packed bed it is necessary to determine the thermophysical properties or their constitutive equations. One of the most uncertain values among them is the solute interaction coefficient of solid particles, which represents the interaction force between solid particles and is equivalent to the viscosity term in Navier-Stokes equations if the dirty fluid model is applied. It was found from the previous study that the solute (solid particle) interaction coefficient, μs, characterizes the solubility such as the melting rate, the distribution of ice volume fraction, the velocity of ice particle, and the melting time. In this study, a parametric study based on the two-dimensional model for the convective melting of granular packed beds (Jiang et al. 1999) is conducted to determine the sensitivity of interaction coefficient to the model prediction. The packed bed considered here is collection of ice particles of various shapes. Warm water at a constant temperature enters horizontally the bed where melting takes place. Two cases are considered. One is to consider μs as constant, and the other is to consider it as a function of the ice volume fraction. The melting rate, fluid flow velocity and ice volume fraction distribution are discussed for different interaction coefficient values. An “optimal” interaction coefficient between ice particles is determined by comparing the simulation data with experimental data (Tao et al. 1998). It is found that the melting results are most sensitive to the value of constant interaction coefficient rather than to whether it is a constant or as a function of the ice volume fraction.