Geomechanical restoration methods are dependent on boundary conditions to ensure geological consistency of the restored model in terms of geometry and strain. Classical restoration boundary conditions, such as flattening a datum horizon, may lead to inconsistent displacement and strain fields. We restore a laboratory structural sandbox model with known deformation history to develop guidelines for definition of boundary conditions that produce improved results from geomechanical restorations. The sandbox model has a basal silicone layer, includes synkinematic deposition, and is characterized by structures analogous to those found in suprasalt extensional environments. The deformed geometry is interpreted from three-dimensional tomography imaging, and a time series of cross section tomography images provides a benchmark to quantify restoration error and inform boundary conditions. We confirm that imposing a lateral displacement equal and opposite to far-field tectonic shortening or extension provides a more accurate restoration. However, the amount of displacement may not be known in real cases. We therefore test several established methods, using only the unrestored geometries, to assess the amount of shortening that should be used to guide geomechanical restorations. An accurate estimation is provided by the area-depth method and potentially by a dilatation analysis. Additionally, novel fault-compliance boundary conditions produce improved results in the vicinity of crossing and branching faults. Application of similar methods should produce improved restoration of natural geologic structures.