The interior of Earth is stratified due to gravity. Therefore, the lateral heterogeneities observed as seismic anomalies by seismologists are extremely interesting: they hold the key to understand the composition, thermal status and evolution of the Earth. This work investigates seismic anomalies inside Earth’s lowermost mantle and focuses on patch-like ultra-low velocity zones (ULVZs) found on Earth’s core-mantle boundary (CMB). Firstly, all previous ULVZ studies are compiled and ULVZ locations on the CMB are digitized. The result is a database, which is publicly available online. A key finding is that there is not a simple mapping between the locations of the observed ULVZs and the large low velocities provinces (LLVPs). Instead, ULVZs are more likely to occur near LLVP boundaries. This spatial correlation study supports a compositionally distinct origin for at least some ULVZs. Next, the seismic structure of the basal mantle beneath the Central America is investigated. This region hosts present and past subducted slabs, which could have brought compositionally distinct oceanic basalt all the way down to the CMB. The waveform distortions of a core-reflected seismic phase and a forward modeling method are used to constrain the causes of the CMB structures. In addition to ULVZ structures, isolated patches of thin zones with shear velocity increased by over 10% relative to background mantle are found for the first time. Ultra-high velocity zones (UHVZs) are interspersed with ULVZs and could be caused by subducted mid-ocean ridge basalt (MORB) that undergoes partial melting and melt segregation. Fe-rich partial melt of MORB can form ULVZs, and silica polymorphs (SiO2) and calcium-perovskite (CaPv) rich solid residue can explain the UHVZs. Finally, large-scale heterogeneities in the lowermost mantle are investigated using S waveform broadening observations. Several basal layer models are case-studied via synthetic calculations. S wave arrivals received at a distance larger than 80˚ in a global dataset from large earthquakes between the years 1994 and 2017 are examined and S waveform broadenings are documented. This approach exploits large distance data for the first time, and therefore is complementary to previous studies in terms of sampling locations. One possible explanation of S waveform broadening is velocity discontinuity inside the D″ layer due to the temperature controlled Bm-pPv phase transition.