Seismic location of tunnels or voids with a cross-borehole survey is examined with field data and theory. The field data were taken at a site with a 2.2-m high by 2.7-m wide, roughly rectangular cross-section tunnel, using a newly developed 1 to 5 kHz system employing a P-wave sparker source. The synthetic records were obtained using a 2.5-D boundary-valued solution for an explosive point source near a cylindrical void, and the solution was evaluated with the method of steepest descent. The synthetic waveforms compared well to the field data; both showed a maximum reduction of amplitude in the tunnel shadow of 8 dB and a maximum first arrival delay of 0.1 ms. Additional theoretical modeling was used to examine the variations of the received signals with tunnel size and frequency and showed amplitude reduction increased with frequency and tunnel size. Calculations showed that S-waves scattered from the tunnel are more than 20 dB smaller than the primary P-wave on hydrophones and more than 12 dB smaller on particle velocity sensors and so could be difficult to see in field data. The close comparison of synthetic waveforms to the field data indicate that the cylindrical model can be used to model data for roughly square cross-section tunnels or voids, as well as for circular cross-section tunnels, and thus is useful for data interpretation and survey planning.