Structures with mechanical joints are difficult to model accurately. Even if the natural frequencies of the system remain essentially constant, the damping introduced by the joints is often observed to change dramatically with amplitude. Although models for individual joints have been employed with some success, accurately modeling a structure with many joints remains a significant obstacle. To this end, Segalman proposed a modal Iwan model, which simplifies the analysis by modeling a system with a linear superposition of weakly-nonlinear, uncoupled single degree-of-freedom systems or modes. Given a simulation model with discrete joints, one can identify the model for each mode by selectively exciting each mode one at a time and observing how the transient response decays. However, in the environment of interest several modes may be excited simultaneously, such as in an experiment when an impulse is applied at a discrete point. In this work, the modal Iwan model framework is assessed numerically to understand how well it captures the dynamic response of typical structures with joints when they are excited with impulsive forces applied at point locations. This is done by comparing the effective natural frequency and modal damping of the uncoupled modal models with those of truth models that include nonlinear modal coupling. These concepts are explored for two structures, a simple spring-mass system and a finite element model of a beam, both of which contain physical Iwan elements to model joint nonlinearity. The results show that modal Iwan models can effectively capture the variations in frequency and damping with amplitude, which, for damping, can increase by as much as two orders of magnitude in the microslip regime. However, even in the microslip regime the accuracy of a modal Iwan model is found to depend on whether the mode in question is dominant in the response; in some cases the effective damping that the uncoupled model predicts is found to be in error by tens of percent. Nonetheless, the modal model captures the response qualitatively and is still far superior to a linear model.