A hybrid reasoning system is developed for damage assessment of structures. The system combines the use of a model of the structure with a knowledge-based reasoning scheme to evaluate if damage is present, its severity (severity and dimension) and its location. Using a given model (or several models), the structural dynamic responses to given excitations are simulated in the presence of different forms of damage. In a ‘learning mode’ an initial casebase is created with the principal features of these damage responses. When the system is working in its operating mode, data acquired by sensors are used to perform a diagnosis by analogy with the cases stored in the casebase, reusing and adapting old situations. Whenever a new situation is detected, it is retained in the casebase to update the available information. This paper describes the methodology and how the system is built and tuned to be ready for operation. This is illustrated by a numerical example of a cantilever truss structure and tested numerically and experimentally with a beam structure. Conclusions are presented with the emphasis on the advantages of using knowledge-based systems for structural assessment.