Computer simulations of the current conduction through the granular mi
crostructure of polycrystalline ZnO varistors are used to explain the
multiple-peaked structure observed in the nonlinearity coefficient mea
sured as a function of voltage for low-voltage varistors and for varis
tors with microelectrode contacts on their surface. Two-dimensional Vo
ronoi networks with random mixtures of electrically nonlinear and ohmi
c elements are used to model the varistors. It is shown that the stati
stical dispersion of the grain sizes has no significant effect on the
nonlinearity coefficient. The multiple-peaked structure is caused by a
small amount of randomly distributed grain boundaries with ohmic beha
vior; i.e., linear current-voltage characteristics of low resistivity.
The observed reduction of the maximum nonlinearity coefficient of pol
ycrystalline varistors, relative to that of isolated grain boundaries,
is attributed to the presence of varistor microjunctions with poor no
nlinearity.