Pounding damage in major earthquakes has been observed frequently in the form of aesthetic, minor or major structural cracks and collapse of buildings. These observations have attracted many numerical and experimental studies that led to analytical models for simulating seismic pounding. This study considers pounding between two steel portal frames without a seismic gap. The first frame has a constant natural period while the second frame has variable stiffness and mass values. Five different ground motions are applied to eight combinations of adjacent frames using a shake table. Numerical simulations for the same configurations are carried out with five pounding force models, viz. linear viscoelastic model, modified linear viscoelastic model, nonlinear viscoelastic model, Hertzdamp model and modified Hertzdamp model. The contact element stiffness and coefficient of restitution for numerical models are determined experimentally. The amplification of maximum displacement of the first frame predicted by the numerical simulations is compared with the shake table results. It was found that the Hertzdamp model always overestimated the responses while the other four models also frequently overestimated the amplifications. The predictions from the four models were not significantly different. Since the linear viscoelastic model requires substantially less computation, compared with the other models this model is more suitable for numerical modelling of pounding responses. However, more study is required to refine the numerical models before building pounding can be modelled with enough confidence.