Epicenters determined for instrumentally observed earthquakes occurring around the turn of the century have figured prominently in estimates of seismic hazard along the Alaska-Aleutian Arc. Unfortunately, because of poor timing control and a sparse station distribution, the locations quoted in the older catalogs can be very unreliable. More reliable epicenter estimates and quantitative assessments of the errors associated with these estimates are needed to address issues related to seismic hazard. Seventeen earthquakes with surface-wave magnitudes above 6.8 that were recorded between 1898 and 1917 are relocated using a modified least-squares criterion. Epicentral estimates are determined using P, S, and S-P arrival times with the a priori constraint that the final location must be near the geographically defined arc. Although epicenters determined using a standard least-squares criterion can be located up to 800 km from the arc, the arc-proximity constraint provides tectonically more meaningful locations, which to a high degree of confidence fit the travel-time observations as well as the unconstrained locations. Epicentral shifts of up to 1,200 km are observed between the relocations and those reported in the earlier catalogs. Seventy-five per cent confidence bounds on the along-arc location of each event vary from 30 to 400 km. Several large events (Ms > 6.9) occurred within or near the rupture zone of the 1965 Rat Islands earthquake and may indicate that this zone failed in a sequence of events around the turn of the century. Of the four events previously suggested as being possible candidates for rupturing either the Shumagin or Unalaska Seismic Gaps, none relocate near these arc segments. The relocation of a large (Ms = 7.9) earthquake which occurred in 1917 within the Shumagin Islands region, however, provides the first strong evidence of largescale seismic moment release within this intensely studied seismic gap.