Engineering structures gradually lose performance over their service life due to damage accumulation. This deterioration is caused by factors such as mechanical and thermal loading, material defects, among others and can result in structural failure. At its initiation, damage is usually associated with a localised loss of stiffness in the affected region and altered structural response. By integrating the current response into a structure’s computer model, one can infer the damage state and prevent failures. Further, on account of model discrepancy and measurement noise, a probabilistic inference is favoured. The aim of this work is to estimate the location and the extent of damage. For this purpose, we use modal response of the structure. It is obtained using Stochastic Subspace Identification (SSI) performed on the acceleration measurements. The damage is represented using parameters of a Finite Element model, an inverse problem is set up and the parameters are estimated using Bayesian inference. Specifically, we define a damage zone with spatially varying damage field that models a reduction in the nominal Young’s modulus within the zone. A set of parameters describe the location and the magnitude of damage in a zone. The number of damage zones is user defined. The salient feature of the work is that we use sparsity promoting priors [1] on damage parameters. They act as a switch indicating the presence or absence of the respective zone. During inference, they mitigate over-parameterisation ensuring that only those zones supported by the modal data are retained. The likelihood function penalises discrepancies between the modal data and the associated model response. We test the approach on experimental data. We use acceleration data from a T-shaped reinforced concrete structure. The structure is progressively damaged through increasing load cycles. After each cycle, the crack locations are documented, and acceleration measurements are taken. The proposed scheme can be readily extended to other failure modes via case-specific parameterisations of damage. [1] Hirsh, Seth M. and Barajas-Solano, David A. and Kutz, J. Nathan, Sparsifying priors for Bayesian uncertainty quantification in model discovery, Royal Society Open Science , Vol. 9, No. 2, 2022.