Random field models in image analysis and spatial statistics usually have l
ocal interactions. They can be simulated by Markov chains which update a si
ngle site at a time. The updating rules typically condition on only a few n
eighboring sites. If we want to approximate the expectation of a bounded fu
nction, can we make better use of the simulations than through the empirica
l estimator? We describe symmetrizations of the empirical estimator which a
re computationally feasible and can lead to considerable variance reduction
. The method is reminiscent of the idea behind generalized von Mises statis
tics. To simplify the exposition, we consider mainly nearest neighbor rando
m fields and the Gibbs sampler.