Attention is rapidly directed to stimuli associated with rewards in past experience, independent of current task goals and physical salience of stimuli. However, despite the robust attentional priority given to reward-associated features, studies often indicate negligible priority towards previously rewarded locations. Here, we propose a relational account of value-driven attention, a mechanism that relies on spatial relationship between items to achieve value-guided selections. In two experiments, participants were trained to associate specific locations with rewards (e.g., high-reward: top-left; low-reward: top-right). They then performed an orientation discrimination task where the target’s absolute location (top-left or top-right) or spatial relationship (“left of” or “right of”) had previously predicted reward. Performance was superior when the target’s spatial relationship matched high-reward than low-reward, irrespective of absolute locations. Conversely, the impact of reward was absent when the target matched the absolute location but not the spatial relationship associated with high reward. Our findings challenge the default assumption of location specificity in value-driven attention, demonstrating a generalizable mechanism that humans adopted to integrate value and spatial information into priority maps for adaptive behavior.