Accurate models of the distribution of elastic heterogeneity in the Ea
rth's mantle are important in many areas of geophysics. The purpose of
this paper is to characterize and compare quantitatively a set of rec
ent three-dimensional models of the elastic structure of the Earth, to
assess their similarities and differences, and to analyze their fit t
o one class of data in order to highlight fruitful directions for futu
re research. The aspherical models considered are the following: M84C
(Woodhouse and Dziewonski, 1984), L02.56 (Dziewonski, 1984), MDLSH (Ta
nimoto, 1990a), SH.10c.17 (Masters et al., 1992), and S12_WM13 (Su et
al., 1994). Through much of the discussion, M84C and L02.56 are combin
ed into a single whole mantle model, M84C + L02.56. The fit of each mo
del to previously tabulated even degree normal mode structure coeffici
ents taken from Smith and Masters (1989a) and Ritzwoller et al. (1988)
for multiplets along the normal mode fundamental and first, second, a
nd fifth overtone branches is also presented. Rather than concentratin
g on detailed comparisons of specific features of the models, analyses
of these models are general and statistical in nature. In particular,
we focus on a comparison of the amplitude and the radial and geograph
ical distribution of heterogeneity in each model and how variations in
each affect the fit to the normal mode observations. In general, the
results of the comparisons between the models are encouraging, especia
lly with respect to the geographical distribution of heterogeneity and
in the fit to the normal mode data sensitive to the upper mantle and
lowermost lower mantle. There remain, however, significant discrepanci
es in amplitude and in the radial distribution of heterogeneity, espec
ially near the top of the upper mantle and near the top of the lower m
antle. The confident use of these models to constrain compositional an
d dynamical information about the mantle will await the resolution of
these discrepancies. The factors that may be responsible for the diffe
rences in the models and/or for the misfit between the observed and pr
edicted normal mode data are divided into two types: intrinsic (or pro
cedural) and extrinsic (or structural). We discuss only three extrinsi
c factors at length here, including errors in the reference crustal mo
dels, unmodeled topography on discontinuities in the interior of the m
antle, and errors in the assumed relationships between shear (upsilon(
s)) and compressional (upsilon(p)) heterogeneity.