The coupled-mode theory (CMT) is a powerful approach routinely used to calculate the effects of spatial mode interactions in perturbed structures, such as optical waveguides. One of its basic hypotheses requires that perturbations are weak. This is usually not the case for devices fabricated with modern semiconductor-based technologies. In this paper, the CMT is studied in these critical cases to assess its validity. Attention will be focused on the quite common case of parallel coupled waveguides. For these structures, results can in fact be compared to the exact ones, obtained using super-modes. The study will show that not all the possible expressions of the coupling coefficients are equivalent, and which one can be pragmatically used to obtain results with minimum errors with respect to exact solutions.