A key challenge in civil engineering is to ensure structural safety and manage the aging infrastructure. In particular, there is an increasing need for more efficient, prompt, and reliable methods for damage detection. In this context, Structural Health Monitoring (SHM) methods play a crucial role. Damage events often manifest as significant variations in the mechanical properties of structural systems, which affect their dynamic response. For this reason, vibration-based SHM methods are particularly suited for damage detection. Typically, classical techniques rely on complex dynamic structural models, which are computationally expensive and, therefore, difficult to use in prompt damage detection tasks. This contribution proposes a data-driven approach to overcome modeling challenges by reframing the damage detection problem as a Bayesian model selection in the quefrency domain. The quefrency domain offers a significant advantage over the frequency domain by enabling a substantial reduction in dimensionality and simplifying the extraction of the quefrency features, i.e., the cepstral coefficients. In this contribution, a Bayesian selection method is employed to choose between two different probabilistic models for cepstral coefficient extraction: (i) an “undamaged model” based on quefrency observations of the structure in its operational state, and (ii) a “damaged model” defined as a general deviation from the “undamaged” one. This Bayesian-based method proves to be highly efficient in damage detection due to its negligible computational cost. Finally, to validate the proposed Bayesian-based damage detection method, a simple numerical application is proposed.