Arrival times from intermediate-depth (110-150 km) earthquakes within the region of flat subduction beneath central Peru provide constraints on the geometry and velocity structure of the subducting Nazca plate. Hypocenters for these events, which are beneath the sub-andean and eastern Peruvian basins, were determined using a best-fitting onedimensional velocity-depth model with a 15-station digitally-recording network deployed in the epicentral region. For that model, P-wave travel times to coastal stations, about 6° trenchward, exhibit negative residuals of up to 4 seconds and have considerably more complexity than arrivals at the network stations. The residuals at coastal stations are conjectured to result from travel paths with long segments in the colder, higher velocity subducting plate. Travel time anomalies were modeled by 3-D raytracing. Computed ray paths show that travel times to coastal stations for the eastern Peru events can be satisfactorilymodeled if velocities relative to the surrounding mantle are 6% lower within the uppermost slab (a 6 km thick layer composed of basaltic oceanic crust) and 8% higher within the cold peridotitic layer (which must be at least 44 km thick). Raytracing runs for this plate model show that "shadow zones" can occur if the source-slab-receiver geometry results in seismic rays passing through regions in which the slab undergoes significant changes in slope. Such geometries exist for seismic waves propagating to some coastal stations from sources located beneath the eastern Peruvian basin. Observed first-arrival times for such cases do in fact have less negative residuals than those for geometries which allow for \direct\ paths. Modeling such arrivals as trapped mode propagation through the high-velocity part of the plate produces arrival times consistent with those observed. / Master of Science