In the current study, we tested whether representations in visual STM (VSTM) can be biased via top–down attentional modulation of visual activity in retinotopically specific locations. We manipulated attention using retrospective cues presented during the retention interval of a VSTM task. Retrospective cues triggered activity in a large-scale network implicated in attentional control and led to retinotopically specific modulation of activity in early visual areas V1–V4. Importantly, shifts of attention during VSTM maintenance were associated with changes in functional connectivity between pFC and retinotopic regions within V4. Our findings provide new insights into top–down control mechanisms that modulate VSTM representations for flexible and goal-directed maintenance of the most relevant memoranda.