This thesis is focused on the properties of microcellular aluminium, produced by replication processing, at the micro, meso and macrostructural level. The replication process is based on the infiltration of a porous sodium chloride (NaCl) perform with molten aluminium. Once the metal is solidified, the resulting composite can be conventionally machined to its final shape. Subsequently leaching away the NaCl in water leads to an open-pore aluminium foam or "sponge", the void size and shape of which can be controlled with wider latitude than in other metal foam production processes. In a first part, the microstructure (i.e. the microstructure of the metal making the foam) of Al-4.5%Cu replicated foams is studied and its influence to the mechanical properties is investigated. These foams solidify following the rules developed for Metal Matrix Composites. The morphology of the microstructure and the microsegregation level are altered due to the confinement of the metal within the NaCl preform. Age hardening is also studied, showing that the classical age-hardening routes developed for bulk Al-4.5%Cu alloys can be applied to Al-4.5%Cu foams, but a dependence of the response to age-hardening to the cell-size is observed, 400 µm cell sizes showing standard hardening while 75 µm cell sizes show no response at low relative density. Alloying the foam brings significant improvements in the mechanical strength of the foams. Mechanical analysis of the foam flow curves shows that the in-situ flow stress of the alloyed metal within 400 µm pore size foam samples depends on the relative density of the foam, and hence on the thickness of struts making the foam. This unexpected result is interpreted as resulting from inhomogeneities in the alloy microstructure, as illustrated by a simple percolation-based model. In a second part, the mesostructure of replicated foams is described with a first-principles geometric model, decomposing the solid into struts and nodes having simplified shapes, modeled using the theory of densifying monosized spherical powder compacts developed by Arzt. The Young's modulus of such foams is estimated based on a beam model, using Timoshenko's beam theory which considers both bending and shear. The model returns a value for the exponent n of the Young's modulus scaling law as a function of relative density that matches experimental data for replicated metal foams, and which significantly exceeds the classical value n = 2, predicted by essentially all preceding beam-theory models. That n exceeds two is caused by changes in the topological features of replicated foams that accompany changes in relative density, and which are captured by the proposed model. Finally, at the macrostructural level, the feasibility and potential of sandwich beams with functionally graded cores by replication processing is demonstrated. Functionally graded foam core beams are produced with up to five foam layers of different densities, distributed across the beam thickness, and tested to show good agreement with analysis. It is also shown that, while lightweight graded metal/metal foam beams show little promise from the standpoint of stiffness-limited design, a longitudinally graded core can generate significant weight savings in yield-limited design when, but only when, there is a gradient in the applied moment along the sandwich beam.