Seismic isolation has been established as an effective earthquake-resistant design method and the lead rubber bearings (LRBs) are among the most commonly used seismic isolation systems. In the scientific literature, a sharp bilinear model is often used for capturing the hysteretic behaviour of the LRBs in the analysis of seismically isolated structures, although the actual behaviour of the LRBs can be more accurately represented utilizing smoothed plasticity, as captured by the Bouc-Wen model. Discrepancies between these two models are quantified in terms of the computed peak relative displacements at the isolation level, as well as the peak inter-storey deflections and the absolute top-floor accelerations, for the case of base-isolated buildings modelled as multi degree-of-freedom systems. Numerical simulations under pulse-like ground motions have been performed to assess the effect of non-linear parameters of the seismic isolation system and characteristics of both the superstructure and the earthquake excitation, on the accuracy of the computed peak structural responses. Through parametric analyses, this paper assesses potential inaccuracies of the computed peak seismic response when the sharp bilinear model is employed for modelling the LRBs instead of the more accurate and smoother Bouc-Wen model.