On several recordings of linear seismometer arrays crossing the Arava Fault (AF) in the Middle East, we see prominent wave trains emerging from in-fault explosions which we interpret as waves being guided by a fault zone related low-velocity layer. The AF is located in the Arava Valley and is considered the principal active fault of the mainly N-S striking Dead Sea Transform System in this section. Observations of these wave trains are confined to certain segments of the receiver lines and occur only for particular shot locations. They exhibit large amplitudes and are almost monochromatic. We model them by a two-dimensional (2-D) analytical solution for the scalar wave field in models with a vertical waveguide embedded in two quarter spaces. A hybrid search scheme combining genetic algorithm and a local random search is employed to explore the multimodal parameter space. Resolution is investigated by synthetic tests. The observations are adequately fit by models with a narrow, only 3--12 m wide waveguide with S wave velocity reduced by 10--60% of the surrounding rock. We relate this vertical low-velocity layer with the damage zone of the AF since the location of receivers observing and of shots generating the guided waves, respectively, match with the surface trace of the fault. The thickness of the damage zone of the AF, at least at shallow depths, seems to be much smaller than in other major fault zones. This could be due to less total slip on this fault.