It remains contentious how information is stored in visual working memory (vWM), even though it is well established that its capacity is limited. Most models have assumed that information is stored in representations that generalise over features. Distinct features have also been assumed to be stored independent of each other. I will present evidence in this thesis that the precision of spatial recall is not only influenced by the number of objects held in memory, but also their location. Moreover, the configuration of the sample array affects recall of single items, suggesting that vWM depends on both, high-resolution local maps and coarse global, representations. Various strategies may be available to overcome the limited capacity of vWM, for example, by utilizing in long-term memory (LTM). I will demonstrate that recall is improved when spatial data are encoding. On the other hand, statistical regularities in the distribution of the memorised objects' locations do not seem to affect recall. Sleep, known to consolidate LTM, has recently been suggested to improve vWM capacity as well. I will present evidence to show that immediate post-learning sleep plays a crutial role in improvingvWM recall by consolidating landmarks in LTM, without improving overall vWM capacity. vWM depends on a distributed cortical network including Visual, Parietal, Frontal and medial Occipito-Temporal areas, though the precise functional role of each area is yet to be established. I will show that binding spatial to visual features, e.g. location and colour, but not binding non-spatial features, is impaired in a patient with bilateral Lingual gyrus and Parahippocampal cortical lesions. Moreover, these lesions disproportionately affect the maintenance of high-resolution spatial data, but not the binding errors, in vWM over longer delays.