A comprehensive data set of selected teleseismic P-wave arrivals and local-network P- and S-wave arrivals from large earthquakes occurring at all depths within a small section of the central Aleutians is used to examine the general problem of earthquake location in island arcs. Reference hypocenters for this special data set are determined for shallow earthquakes from local-network data and for deep earthquakes from combined local and teleseismic data by joint inversion for structure and location. The high-velocity lithospheric slab beneath the central Aleutians may displace hypocenters that are located using spherically symmetric Earth models; the amount of displacement depends on the position of the earthquakes with respect to the slab and on whether local or teleseismic data are used to locate the earthquakes. Hypocenters for trench and intermediate-depth events appear to be minimally biased by the effects of slab structure on rays to teleseismic stations. However, locations of intermediate-depth events based on only local data are systematically displaced southwards, the magnitude of the displacement being proportional to depth. Shallow-focus events along the main thrust zone, although well located using only local-network data, are severely shifted northwards and deeper, with displacements as large as 50 km, by slab effects on teleseismic travel times. Hypocenters determined by a method that utilizes seismic ray tracing through a three-dimensional velocity model of the subduction zone, derived by thermal modeling, are compared to results obtained by the method of joint hypocenter determination (JHD) that formally assumes a laterally homogeneous velocity model over the source region and treats all raypath anomalies as constant station corrections to the travel-time curve. The ray-tracing method has the theoretical advantage that it accounts for variations in travel-time anomalies within a group of events distributed over a sizable region of a dipping, high-velocity lithospheric slab. In application, JHD has the practical advantage that it does not require the specification of a theoretical velocity model for the slab. Considering earthquakes within a 260 km long by 60 km wide section of the Aleutian main thrust zone, our results suggest that the theoretical velocity structure of the slab is presently not sufficiently well known that accurate locations can be obtained independently of locally recorded data. Using a locally recorded earthquake as a calibration event, JHD gave excellent results over the entire section of the main thrust zone here studied, without showing a strong effect that might be attributed to spatially varying source-station anomalies. We also calibrated the ray-tracing method using locally recorded data and obtained results generally similar to those obtained by JHD. ?? 1982.