Controlling failure modes is an important issue for the development of bonded assemblies of multi-material structures in industry. The analysis of the failure modes of bonded assemblies, and in particular the evaluation of the adhesive/cohesive failure rate, constitute a significant issue. Generally, it is considered in industrial applications that obtaining a cohesive fracture makes it possible to generate better predictions regarding the rupture of the adhesive joints, even if recent work has shown that a good prediction of the mechanical behavior can also be obtained when adhesive or mixed failure modes occur. Currently, these evaluations are carried out by experts by means of a visual analysis, either directly on the fracture surfaces, or sometimes from high-resolution microscopy. The approach presented in this paper aims to set out a methodology based on the Reflectance Transformation Imaging technique for the objective characterization of experimental fracture surfaces of bonded assemblies. We show how this imaging technique can be used to quantify objective surface features, and how these local descriptors can be used to estimate and map out the failure mode on the fracture surfaces.