The scattering properties of tetrahedrally bonded amorphous materials are broadly consistent with a random network structure. Evidence obtained by electron microscopy appears to contradict this conclusion, and to favour a microcrystalline structure. The conflict of evidence is not resolved by the discovery that a particular random network model can produce lattice fringes in a computed electron micrograph, since the probability of this occurrence is too low to account for what has been observed in practice. After reviewing recent work on this topic, we conclude that the fringes may be an artefact introduced by insufficient stability of the parameters which determine lens aberrations. Further experimental work is needed.