High-quality undisturbed samples of fluvial sand were obtained from the field using the ground freezing technique. In the laboratory, the in situ void ratio of these high-quality undisturbed frozen samples was replicated using four different reconstitution methods: dry deposition, moist tamping, water sedimentation (by spooning) and slurry deposition. The initial fabric of the specimens was evaluated using X-ray micro-computed tomography and advanced image analysis. Initial fabric features were assessed in terms of the particle orientation, anisotropy, void ratio distribution and particle sizes within the specimens. Fabric analysis results suggest that none of the laboratory reconstitution techniques used captures the true three-dimensional initial fabric of undisturbed fluvial sand. However, the slurry deposition method managed to reproduce the inherent particle orientation, anisotropy and the variations of void ratio and particle sizes of the undisturbed fluvial sand. This observation explains why previous rigorous studies on the macro-mechanical behaviour of sands deposited under water have systematically demonstrated that slurry deposition is the most suitable method to reconstitute in the laboratory natural sands deposited under water. This has major implications for geotechnical testing and analyses of liquefaction of sands deposited under water such as fluvial, offshore and tailings sands.