Surface nuclear magnetic resonance (surface NMR) has up to now rarely been applied to 3D subsurface modeling. Inversion approaches currently in use are smooth inversion techniques that are not useful for identifying sharp geologic boundaries. Although they are already computationally expensive, the resulting models are restricted to imaging the subsurface water content distribution and do not deliver relaxation times based on the QT inversion scheme established elsewhere. We have developed a method of 3D block QT inversion that uses horizontal smoothness constraints to resolve sharp boundaries in the vertical direction and the distributions of the water content and relaxation time . We have improved the computational efficiency, i.e., the ability to perform the inversion using a common desktop computer, by gating the surface NMR data, reducing the model space to monoexponential decays within the subsurface bodies, and inverting based on blocklike structures instead of smooth distributions. We have developed a synthetic study to assess the effectiveness of our block QT inversion technique in imaging 3D water content distributions, and we compared the results with those of a smooth inversion. Furthermore, we evaluated results from a field survey conducted on the frozen surface of an artificial lake. We found that our block QT inversion approach provides results that are superior to those of smooth inversion and consistent with the available construction plan of the lake. We expect that 3D block QT inversion will be a useful approach also in other geologic settings, such as buried valleys, because it overcomes the current limitations of applying 3D surface NMR inversion.