When can shape and scale parameters of a 3D variogram be estimated?

We have used a method of least squares to fit full 3D variogram models to data, and have tested it on data taken from Gaussian random fields. The empirical variogram estimates are made using various lag grid definitions and the best of these grids is identified. Our results suggest that some 200 vertical wells are needed for obtaining reliable estimates of the azimuth and dip anisotropy angles, while some 50 wells seem sufficient for the horizontal ranges and the sill. For the vertical range 10 wells are sufficient.