Recent studies of the magnetic field at the core--mantle boundary have revealed fixed sites of either static magnetic features or persistent secular variation. This suggests that part of the magnetic-field behaviour is controlled by the mantle. The most plausible mechanism for core--mantle interaction is thermal coupling, although topography may also be significant. The magnetic sites coincide with anomalies in lower-mantle seismic velocity, as determined from tomography, and density, as determined by flow models of mantle convection constrained by tomography and the geoid. Some magnetic features coincide with subduction zones, particularly those near Indonesia; they may be caused by bumps on the core--mantle boundary beneath trenches. Palaeomagnetic pole positions suggest the magnetic behaviour has persisted for at least 5 Ma, as would be expected if it were controlled from the mantle. These conclusions could be quantified if the frozen-flux hypothesis allowed determination of fluid flow at the core surface, but unfortunately failure of the hypothesis makes all such determinations suspect. Core motions calculated so far suggest the flow is mainly toroidal. Questions about the dynamics of the flow (whether it is steady, stratified, or geostrophic) remain unresolved.