It has recently been proposed that all deep earthquakes (>400 km depth) are caused by a shear instability associated with the transformation of metastable olivine to its high-pressure polymorphs within the cold core of subducting slabs (Green and Burnley, 1989; Kirby et al., 1991). If so, then the seismogenic zone would narrow with depth below 400 km following slab isotherms. Because the ratio of fault slip to length does not increase markedly with depth, and stress orientations indicate that deep fault planes are not parallel to the subducting slab, a fault plane large enough to generate a very large, deep earthquake would not fit within a narrow seismogenic zone and multiple fault planes would be needed. If the fault planes are not parallel due to spatial variations in stress, a deviatoric non-double-couple component (volume-preserving, non-shearing part) of the seismic moment tensor is likely to result. Thus, if the seismogenic zone narrows with depth, one would expect an increasing nondouble-couple component (1) for large earthquakes as depth increases from 400 to 700 km, and (2) for very deep earthquakes (below 550 km depth) as seismic moment increases. Data from the Harvard CMT catalog, augmented by moment tensor inversions for the two largest deep earthquakes during the past thirty years, are examined for these features. Although significant scatter is present, the data roughly follow the trends described above, suggesting a relationship between the non-double-couple component of deep earthquakes and the geometry of a seismogenic zone narrowing with depth. ¿ American Geophysical Union 1993