Following Hasegawa and Sacks (1981), the Benioff zone geometry beneath southern Peru is determined using 32 months of arrival-time data from a local seismic network. Various earthquake location algorithms a re-tested by comparing the locational estimates and error statistics produced by each, to determine if any one method produces solutions which are more stable than the others. No significant differences were found among three of the four methods considered for this data set. We then compare the epicentral and depth confidence intervals produced by these algorithms using the method described by Evernden (1969), and find that the three-parameter method produces confidence regions which are smaller and require more assumptions than the four-parameter method's. Therefore, the four-parameter method, including both P and S arrivals was used in this study. Our data set includes 2476 located events, of which 205 are chosen to be master events; earthquakes with the most reliable locations. All the events are then relocated using source- region dependent station corrections derived from the station residuals of the master events. We find that the Benioff zone in the more southerly region of our study area dips at a nearly constant 30°, while in the more northerly region this trend is apparent only down to a depth of 100 Km., at which point the Benioff zone becomes nearly horizontal. By analyzing the spatial aspects of our data set between these two regions, we infer that the deformation of the Benioff zone is continuous; i.e., there is no discontinuity in the seismicity that might suggest a tear in the subducting plate. / M.S.