According to Kubota et al (here), three of the main problems are:
1) Thickness: Accurately determining the thickness of serial sections. Messing this up will introduce error into 3d reconstructions. To get around this people use the minimal folds method. This looks for protrusions in the plane where small folds of tissue self-adhere. Section thickness is assumed to be one-half the width of these protrusions. However, in reality this method can give very variable results. Kubota et al suggest an “optical method” that uses a 3D laser confocal microscope instead, which gives much more reliable results, at least under resin-based conditions:
2) Synapse identification: Often synaptic junctions parallel to the section plane (or at a low angle to it) are not counted as synapses. This occurs ~ 25% of the time! So the authors use a sequence of 3d section analysis that accurately predicts the presence of a synapse, which you can see in the picture below: “(1) many synapse vesicles (A / B below), (2) presynaptic grids (C), (3) synaptic cleft structures, (4) postsynaptic densities (D), and (5) cytoplasm of postsynaptic dendrite or spine (E)”:
3) Shrinkage: They note that in their previous studies, after fixation, dehydration, and embedding, their GABAergic nonpyramidal cells often shrunk to up to 90% of their original size. They offer no solutions for this problem.
Elsewhere, Cordona et al note that “reconstruction of microcircuits requires serial section electron microscopy, due to the small size of terminal neuronal processes and their synaptic contacts. Because of the amount of labor that traditionally comes with this approach, very little is known about microcircuitry in brains across the animal kingdom. Many of the problems of serial electron microscopy reconstruction are now solvable with digital image recording and specialized software for both image acquisition and postprocessing.”
Kubota Y, Hatada S and Kawaguchi Y (2009) Important factors for the three-dimensional reconstruction of neuronal structures from serial ultrathin sections. Front. Neural Circuits 3:4. doi:10.3389/neuro.04.004.2009