Seismic rays are traced through a prescribed three-dimensional inhomogeneity that simulates the subducted slab below the Shumagin Islands region to calculate local station delays for a given hypocenter and a slab model. Hypocenters determined by the Shumagin seismic network are then relocated using the station delays, a flat-layered velocity structure and a standard earthquake location computer program. Station delays are calculated for 12 hypocenters with respect to six different slab models to identify the slab model that is most consistent with the available arrival time and waveform data. A set of path corrections that is calculated for each grid point-station pair on a preliminary grid of 36 points in the depth range from 60 to 30 km is used to recalculate the hypocenters for all of the 1982 earthquakes with depths greater than 50 km. Application of this method to data from 1982 or the digitally recording Shumagin seismic network shows the following results: (1) a previously observed apparent increase in dip of the subducted slab at depths of ≈100 km disappears, (2) the subducted slab can be modeled as a dipping structure that dips at a constant angle of 45¿ toward north-northwest at depths between 80 and 250--300 km and has a 7% higher velocity than the surrounding mantle, (3) hypocenters determined from Shumagin network data are located only 10--20 km south of high-quality hypocenters determined from teleseismic data alone, (4) qualitative comparison of digitally recorded seismograms with calculated ray paths shows enrichment of high-frequency coda, possible converted phases, and low amplitudes of first P arrivals for rays that travel mostly along the slab. Conversely, for rays that travel almost vertically through the upper plate the seismograms show a high amplitude of first P arrivals that are followed by an insignificant coda and low-amplitude S waves.