Nuclear explosions at the French test sites in the South Pacific have been recorded at the regional seismic networks in the western United States. A total of approximately 600 good-quality, short-period seismograms were selected from an initial set of 900. The dense receiver spacing, particularly in California, allows the division of the networks into numerous subarrays and the data from each subarray can be analysed using conventional array techniques such as beam-forming and frequency—wave-number analysis. The seismic rays provide dense sampling of a small area of the core-mantle boundary (CMB) beneath the eastern Pacific from 5° to 13°N and 133° to 126°W. PcP has a shallow incidence angle on the CMB at an epicentral distance of 65° and D″ is sampled over a large swathe from 2°S to 24°N and from 134° to 123°W. Beam traces have been formed for the P and PcP phases for each of the subarrays, using slowness and azimuth values derived from array methods. Good-quality beams have been selected for further analysis. By applying a matched filter to the beamed traces, an accurate set of PcP-P differential travel times has been determined and compared with IASP91. The residuals show a smooth trend over the area sampled. For the given epicentral distances, P and PcP have similar paths through the crust and upper mantle so we make the assumption that near-source and near-receiver structures have negligible effect on the differential times. We also assume that the midmantle is relatively homogenous. Many previous studies have interpreted PcP-P residuals in terms of CMB topography. By mapping the residuals directly to topography, a range in elevation of 40 km over a lateral extent of about 700 km results. Alternatively, we attempted to interpret the residuals in terms of velocity heterogeneity in D″. A range of D″ models have been tested including lateral velocity variation in a D″ layer with constant thickness. The PcP-P residuals may be explained by allowing the P-wave velocity to vary laterally by a factor of 2% over the region sampled, with most of the region being marginally fast. This suggests that D″ heterogeneity has large spectral power at relatively short wavelengths. In this scenario, previous tomographic inversions for CMB topography will have large errors since in these inversions, corrections for D″ structure are based on low-order spherical harmonic models.