Inertial sub-range statistics of stochastic models for the Lagrangian
velocity in turbulent flow have been examined. For Markovian models it
is shown that consistency with Kolmogorov's theory of local isotropy
requires that the Lagrangian velocity be a continuous function of time
. This limits suitable Markov models to those which can be represented
by a stochastic differential equation. Markov models in which the vel
ocity is discontinuous (and a class of non-Markovian jump models) are
not consistent with Kolmogorov's theory. Modifications to Kolmogorov's
theory to account for the effects of intermittency are shown to be no
n-Markovian, but still correspond to a Lagrangian velocity which is co
ntinuous. In Gaussian homogeneous turbulence only continuous Markov mo
dels predict that the particle displacement is Gaussian. For a Markovi
an jump model, the particle displacement distribution is leptokurtic w
ith a maximum excess of about 0.67, which is inconsistent with wind tu
nnel data in grid turbulence.