SUMMARYSeismic anisotropy has been detected at many depths of the Earth, including its upper layers, the lowermost mantle and the inner core. While upper mantle seismic anisotropy is relatively straightforward to resolve, lowermost mantle anisotropy has proven to be more complicated to measure. Due to their long, horizontal ray paths along the core–mantle boundary (CMB), S waves diffracted along the CMB (Sdiff) are potentially strongly influenced by lowermost mantle anisotropy. Sdiff waves can be recorded over a large epicentral distance range and thus sample the lowermost mantle everywhere around the globe. Sdiff therefore represents a promising phase for studying lowermost mantle anisotropy; however, previous studies have pointed out some difficulties with the interpretation of differential SHdiff–SVdiff traveltimes in terms of seismic anisotropy. Here, we provide a new, comprehensive assessment of the usability of Sdiff waves to infer lowermost mantle anisotropy. Using both axisymmetric and fully 3-D global wavefield simulations, we show that there are cases in which Sdiff can reliably detect and characterize deep mantle anisotropy when measuring traditional splitting parameters (as opposed to differential traveltimes). First, we analyze isotropic effects on Sdiff polarizations, including the influence of realistic velocity structure (such as 3-D velocity heterogeneity and ultra-low velocity zones), the character of the lowermost mantle velocity gradient, mantle attenuation structure, and Earth’s Coriolis force. Secondly, we evaluate effects of seismic anisotropy in both the upper and the lowermost mantle on SHdiff waves. In particular, we investigate how SHdiff waves are split by seismic anisotropy in the upper mantle near the source and how this anisotropic signature propagates to the receiver for a variety of lowermost mantle models. We demonstrate that, in particular and predictable cases, anisotropy leads to Sdiff splitting that can be clearly distinguished from other waveform effects. These results enable us to lay out a strategy for the analysis of Sdiff splitting due to anisotropy at the base of the mantle, which includes steps to help avoid potential pitfalls, with attention paid to the initial polarization of Sdiff and the influence of source-side anisotropy. We demonstrate our Sdiff splitting method using three earthquakes that occurred beneath the Celebes Sea, measured at many transportable array stations at a suitable epicentral distance. We resolve consistent and well-constrained Sdiff splitting parameters due to lowermost mantle anisotropy beneath the northeastern Pacific Ocean.