The aim of this paper is to establish a criterion able to distinguish cohesive and adhesive defects in structural joints. To do this, two ultrasonic rheological models are compared to a trilayer model: (1) a cohesive rheological model that corresponds to the standard approach, i.e., the entire adhesive layer is replaced by a spring surface distribution with inertia, and (2) an adhesive rheological model, which consists in describing only the contact zones by a spring surface distribution. It then appears that the cutoff frequencies of guided modes give additional information on the quality of joining. Indeed, the cohesive rheological model makes it possible to quantify global (cohesive) defects in the adhesive layer, while the adhesive rheological model targets local (adhesive) defects on the interface level. To validate this result, ultrasonic experiments and mechanical tests are successively carried out on two types of samples devoid of cohesion defects. Model (2) gives a good description of the differences observed on samples having undergone or not undergone a surface pretreatment.