In the framework of Structural Health Monitoring, vibration-based methods are commonly used to assess the condition of a structural system, being the dynamic properties sensitive to damage-induced changes. Within this context, negative selection, a bio-inspired classification algorithm, can be exploited to distinguish anomalous from normal behaviours by comparing the monitored system features with a set of detectors appropriately trained to spot any possible anomaly inside the unitary feature space. Such method results particularly convenient due to its easy implementation, low computational cost and capability to carry out the classification based on a training set of data belonging only to a healthy-state condition. This circumstance is extremely common in real civil engineering applications where no knowledge might exist about different structural conditions over time. In this paper, a negative-selection algorithm with a non-random strategy for detector generation is developed and tested on a numerical case study, namely a model simulating the I-40 Bridge over the Rio Grande in Albuquerque, New Mexico (USA). The work carried out proves that the algorithm is suitable for the purpose of damage detection (a binary classification problem) and, by introducing the anomaly score as a qualitative measure of the level of damage, provides a sound analysis of the method multiclass classification skills, aiming at the quantification of the damage.