To prepare to update location procedures at the International Seismological Centre (ISC), we conduct a global test of location accuracy using two 3D Earth models and three 1D models (Jeffreys-Bullens [J-B], Preliminary Reference Earth Model [PREM], ak135) with initial phase arrival times from different distance ranges. The 3D models include one spherical harmonic model and one irregular grid model, representing two important alternative model parameterizations. Our tests include a worldwide distribution of reference events from a previous study by Smith and Ekström (1996) and the ground-truth database of the prototype International Data Centre. For each test we use phase arrival times associated with the events in the ISC Bulletin and recompute hypocenters with the same algorithm used in the ISC operations. With teleseismic arrival times alone (Δ ≥ 25°), the overall accuracy of our recomputed hypocenters is similar using either of the 3D models. The teleseismic hypocenters from 3D models are, on an average, closer to the reference locations than the hypocenters computed using any of the 1D models; but the improvement in results using ak135 is small. When we use the irregular grid model with teleseismic, regional, and local initial arrival times together and fix to reference depths, the accuracy of the epicenters is improved over the results with teleseismic times alone. With a 3D model we are able to invert for depth for a larger fraction of earthquakes, and our newly computed depths are more tightly clustered in a subduction zone than depths from the ISC Bulletin, but not as tightly as depths computed by Engdahl et al. (1998) who used depth phases as well as direct phases. In a test of all the locatable events from 1 month of the ISC Bulletin the irregular grid model improved residuals and formal location uncertainties without introducing large systematic changes in the hypocenters.