Built-up structures, such as airplanes, ships, and even refrigeration systems, which have many components, can be substructured to speed up and facilitate the process of calculating the vibratory response of the complete system. In many structures, there are rubber isolators that connect component parts, and these connections can each occur over a finite distributed area. It is often convenient and intuitive to substructure the system at the isolators. However, in previous work, it has been shown that the frequency response of the complete system does not always agree with the frequency response of the system calculated from the mobilities of the subsystems. It was thought that this was due to the distributed area connection of the isolators, and this motivated the study reported in this article. An investigation into some issues that occur when substructuring a system that contains soft distributed isolators is described. Using finite element models, it is shown that if a system is substructured, such that the interface between the substructures occurs at a soft rubber isolator, then there is a limited frequency range over which the frequency response function of the assembled system is accurate. It is further shown that it is far better to substructure the system, at stiff, discrete connections, if possible. The frequency range over which the frequency response of the assembled system should then be more accurate over a much wider frequency range.