In March 1998 a small-aperture (∼350 m) linear array of five broadband stations was installed across a Pliocene strike-slip fault crossing the town of Nocera Umbra, central Italy. Three-component seismograms of local earthquakes recorded during the Umbria-Marche seismic sequence have shown that the degradation of mechanical properties in damaged rock causes a significant amplification of ground motion within the fault zone (Marra et al., 2000). During the operation of the array, an MW 5.3 subcrustal earthquake occurred at a depth of 48 km and an epicentral distance of 10 km from Nocera Umbra. The near-vertical incidence of this event offers a unique opportunity for constraining the 2D geometry of the laterally heterogeneous low-velocity structure responsible for the large amplification across the fault. Based on the observed symmetry of the amplification pattern, a 2D wedge-shaped model of the fault zone is used for the computation of antiplane (SH) synthetic waveforms to be compared with observations along the array. Two parameters of the wedge model can be constrained by independent geophysical and geological data: (1) shear-wave velocity in pristine rock outside the fault zone (vS = 2200 m/sec) and (2) the lateral extent of the damaged rock zone (L = 160 m). The other two model parameters, namely, the wedge depth (h) and shear-wave velocity of the fault zone (vFS), are varied. Computer simulations show that trapped waves, which have been observed so far for earthquakes occurring inside the fault zone, can develop inside a 2D wedge model when struck by external, vertically incident seismic radiation. Results of numerical modeling indicate that, in spite of its extreme simplicity, such an idealized model is successful in reproducing many of the essential characters of ground motion within the fault zone of Nocera Umbra. Even though the best-fit solution is not unique and the model parameters trade off each other, a study of parameter sensitivity, comparing synthetics with recorded data, indicates that the variability of h and vFS can be restricted to a limited range of values: the quality of fit is optimal when the shear-velocity reduction lies in the range of 40%–50% and the depth of the trapping zone is comprised between 1 and 2 km.