Detailed knowledge about the three-dimensional morphology of the human cochlea and its intra-cochlear bony and soft-tissue structures is essential for development of new cochlear implant electrode carriers. A manual cross-sectional preparation and imaging technique, hereinafter referred to as “microgrinding”, uses human temporal bone samples embedded in epoxy resin. This process was automated to shorten the time needed for preparation and to increase reproducibility. In this study, reconstruction accuracy of the automated microgrinding technique was determined. Four assemblies of LEGO® bricks were used as artificial samples to analyze the resulting reconstruction accuracy of the whole procedure including embedding, preparation and image registration. The outer surfaces of the samples were measured using a portable coordinate measuring machine by manually choosing points on each surface. After embedding the samples in epoxy resin, following the protocol for human temporal bone samples, preparation using the automated microgrinding was performed with a slice thickness of 50μm. Pixel-spacing within an image was 11.5μm/px. The samples were identified within the dataset using threshold segmentation. Subsequently, points were manually chosen on each surface of the segmented samples. Planes, angles between planes and distances between corners were calculated for the physical and the digital samples. Deviation of the digitally derived measures compared to the physically derived ones describe the reconstruction accuracy of the automated microgrinding process. Parallel planes had a deviation of 0.5° ± 0.5° (range: 0 - 2.9°). Orthogonal planes, whose line of intersection is parallel to the direction of milled abrasion, had a deviation of 0.8 ° ± 0.5 ° (range: 0 - 2.1°), whereas such orthogonal planes whose line of intersection is parallel to the documented sample surface had a deviation of 0.4° ± 0.3 ° (range: 0 - 1.7 °). Distances between corners had an absolute deviation of 0.1 mm ± 0.1mm (range: 0 - 0.6mm), which corresponds to a relative deviation of 1% ± 1% (range: 0 - 6.5%). This study shows a highly accurate sample preparation using an automated microgrinding technique as an essential prerequisite for three-dimensional morphological imaging.