We stack long‐period, transverse‐component seismograms recorded by the Global Digital Seismograph Network (GDSN) (1976–1996), Incorporated Research Institutions for Seismology‐International Deployment of Accelerometers (IRIS‐IDA) (1988–1996), and Geoscope (1988–1996) networks to map large‐scale topography on the 410‐ and 660‐km seismic velocity discontinuities. Underside reflections from these discontinuities arrive as precursors to the SS phase, and their timing can be used to obtain global variations of the depth to the reflectors. We analyze over 13,000 records from events mb>5.5, focal depth <75 km, and range 110° to 180° by picking and aligning on SS, then stacking the records along the theoretical travel time curves for the discontinuity reflections. Separate stacks are obtained for 416 equally spaced caps of 10° radius; clear 410‐ and 660‐km reflections are visible for almost all of the caps while 520‐km reflections are seen in about half of the caps. The differential travel times between the precursors and the SS arrival are measured on each stack, with uncertainty estimates obtained using a bootstrap resampling method. We then compute discontinuity depths relative to the isotropic Preliminary Reference Earth Model (PREM) at 40‐s period, correcting for surface topography and crustal thickness variations using the CRUST5.0 model of Mooney et al. [1995], and for upper mantle S velocity heterogeneity using model S16B30 of Masters et al. [1996]. The resulting maps of discontinuity topography have more complete coverage than previous studies; observed depths are highly correlated between adjacent caps and appear dominated by large‐scale topography variations. The 660‐km discontinuity exhibits peak‐to‐peak topography of about 38 km, with regional depressions that correlate with areas of current and past subduction around the Pacific Ocean. Large‐scale topography on the 410‐km discontinuity is lower in amplitude and largely uncorrelated with the topography on the 660‐km interface. The width of the transition zone, WTZ, as measured by the separation between the 410‐ and 660‐km discontinuities, appears thickest in areas of active subduction (e.g., Kurils, Philippines, and Tonga) and thins beneath Antarctica and much of the central Pacific Ocean. Spatial variations in WTZ appear unrelated to ocean‐continent differences but do roughly correlate with the S16B30 velocities in the transition zone, consistent with a common thermal origin for both patterns. The lower‐amplitude 520‐km reflector is more difficult to resolve but appears to be a global feature as it is observed preferentially for those bounce point caps with the most data.