Third generation wave-power devices are usually envisaged as being either a single large device or an array of smaller devices. The benefit of an array, compared to a single device, is that the individual components are relatively inexpensive to repair and replace; however issues arise due to interaction between the array members, which can lead to constructive or destructive interference of the wave-field, thus increasing or decreasing the power that can be absorbed. This thesis is concerned with the optimal formation and design of these arrays of wave-power devices from a hydrodynamic perspective. Previous literature has indicated that a deterministic optimisation of the array layout, which directly maximises the array performance, results in high sensitivity of the optimal performance to incoming wave parameters. This work considers a more robust optimisation, where the mean performance of the array is maximised. Determining the optimal array configuration is associated with numerical optimisation. Previous studies have shown that a balance must be struck between accurately modelling the devices of the array (including their interactions) and the requirement of establishing a reliable optimisation process. Thus, linear wave theory and the point absorber approximation are utilised within this work. Several array geometries are investigated, including linear and circular arrays, along with a general 2D optimisation without any imposed symmetry. Both constrained and unconstrained WEC motions are considered. Regular waves are assumed for the majority of this work, with a preliminary extension to irregular waves also investigated for elementary linear arrays. In general, it is shown that optimal unconstrained arrays tend to contain closely spaced groups of WECs, while constrained arrays are more spread out. A trade-off between peak performance and performance stability is identified for general WEC arrays, while linear arrays also exhibit a trade-off between stability to wavenumber variations and incident wave angle variations. Overall, it is shown that linear arrays perform poorly for some orientations, regardless of the array layout. Better constructive interaction can be achieved in beam seas for unconstrained motions, while head seas allow for the best interaction when WEC motions constraints are applied. As expected, better interaction can be achieved for more general array layouts, without a prescribed geometry.