The problem of recovery of three-dimensional microstructural parameters from two-dimensional images is addressed in its inverse form. Three-dimensional inhomogeneous solids with known microstructures are generated virtually and images of their cross-sections are analyzed. Three different types of randomly oriented microcracks and microcracks having preferential orientation with some scatter. Somewhat unexpected result is that, in all the cases considered, three-dimensional defect densities are very close to average values of corresponding two-dimensional parameters.