Random realizations of three-dimensional exponentially correlated hydr
aulic conductivity fields are used in a finite-difference numerical fl
ow model to calculate the mean and covariance of the corresponding Lag
rangian-velocity fields. The dispersivity of the porous medium is then
determined from the Lagrangian-velocity statistics using the Taylor d
efinition. This estimation procedure is exact, except for numerical er
rors, and the results are used to assess the accuracy of various first
-order dispersion theories in both isotropic and anisotropic porous me
dia. The results show that the Dagan theory is by far the most robust
in both isotropic and anisotropic media, producing accurate values of
the principal dispersivity components for sigma gamma as high as 1.0.
In the case of anisotropic media where the flow is at an angle to the
principal axis of hydraulic conductivity, it is shown that the dispers
ivity tensor is rotated away from the how direction in the non-Fickian
phase, but eventually coincides with the flow direction in the Fickia
n phase.