I've started reading up on structural neurobiology recently. Ragan et al., 2012 describes a two-photon microscopy technique that can slice and scan a mouse brain at a resolution of 2 μm^3 in a week. Eight years earlier Denk and Horstmann (2004) presented a similarly automated method for slicing and scanning neural tissue in an electron microscope (EM).
EM resolutions of about 20 nm^3 allow us to model the complex shapes of individual neurons and synapses. The only brain that has in fact been completely reconstructed in this way though is the tiny nervous system (~300 neurons) of C. Elegans (White et al., 1986). This is because reconstructing cell-shapes in 3D, neuron by neuron, from thousands and thousands of grainy images is a very tedious yet difficult job. One solution is to have multiple amateurs work on the same EM data and average their results (Helmstaedter et al., 2011). Another is to train computers to analyse the EM data for us (Jain et al., 2012).
There are many reasons for developing high-resolution 3D models of brains (Denk et al., 2012; Lichtman and Sanes, 2008). If we want one day to have computer systems that replicate all the important functions of biological brains we need understand when and why anatomical detail matters. What if for example we assume that all axons and dendrites in a model are cylindrical, or that all synapses have the same geometry; would such simplifications severely disrupt network function? Seems the high-throughput tools we need to answer such questions are only just being made available.