A FRF-based substructuring method attempts to predict the dynamic characteristics of a complex structure from predetermined FRFs of the comprising uncoupled substructures. Although this method has the advantage of being able to incorporate experimental component FRFs directly, it is prone to errors : measurement errors, coordinate incompleteness, modal incompleteness, etc. Among the various sources of errors, this paper deals with the problem of modal incompleteness (or residual problem) of which importance is underestimated compared to others. It is a well-known rule of thumb that such a problem can be overcome by including modes up to 2 or 3 times the upper frequency of interest. Using a simulated case study, it is demonstrated that even including modes up to 20 times the upper frequency of interest does not guarantee a satisfactory result. A method to compensate the residual errors is introduced. This method requires the whole FRF matrices of substructures which is practically impossible for a complex structure. An applicable alternative is suggested and applied successfully to the case study. Finally, the effects of measurement errors on the residual compensation are also discussed.