The understanding of mantle convection is one of the most puzzing problems of modern geophysics. Among the different approaches used by geophysicists to investigate mantle convection, seismic tomography is the only one able to visualize, at the same time, temperature, petrological anomalies, and flow directions from seismic velocity and anisotropy heterogeneities. In order to enable a comparison with other geophysical observables, tomographic models are expanded into spherical harmonics. most tomographic models agree that down to 300--400 km, deep structure is closely related to plate tectonics and continental distribution. Its corresponding spectral content regularly decreases with decreasing wavelength. At greater depths in the transition zone, degree 2 and (to a lesser extent) degree 6 distributions become predominant. A degree 2 pattern is also present in the lower mantle and is strongly correlated with the geoid but offset with respect to the degree 2 pattern of the transition zone.

A simple flow pattern with two upgoing and two down-going large-scale flows can be invoked to simply explain the predominance of degrees 2 and 6 for seismic velocity and degree 4 for radial anisotropy. Therefore below the apparent complexity of plate tectonics, it turns out that mantle convection is surprisingly simply organized in the transition zone. Between 400 and 1000 km, these large-scale flows are not independent from the circulation in the first 400 km but are connected to some of the most tectonically active zones (fast ridges and slabs). It is also suggested that the degree 6 which seems to be a marker of the hotspot distribution is not independent of the deep degree 2 but might be the consequence of this simple flow pattern. The good correlation between seismic tomography degree 6 and hotspots degree 6 favors an origin at depth of hotspots in the transition zone. Generally, the mantle cannot be divided into independent convecting cells but is characterized by imbricated convection, where different scales coexist and where exchange of matter is possible. Therefore seismic tomography is able to provide very strong constraints on possible models of mantle convection, but many features are still unexplained. Only very long spatial wavelengths are well resolved so far, and a complete understanding of mantle dynamics necessitates relating the different scales present in convective processes. ¿ American Geophysical Union 1994