Visual attention prioritizes information presented at particular spatial locations. These locations can be defined in reference frames centered on the environment or on the viewer. This study investigates whether incidentally learned attention uses a viewer-centered or environment-centered reference frame. Participants conducted visual search on a monitor laid flat on a tabletop. During training, the target was more likely to appear in a "rich" quadrant than in other "sparse" quadrants. Although participants were unaware of this manipulation, they found the target faster in the rich quadrant than in the sparse quadrants, showing probability cuing. In a subsequent testing phase, participants were reseated to change their viewpoint by 90 degrees. In addition, the target became equally likely to appear in any quadrant. Spatial attention continued to be biased for several hundred trials. Critically, the attentional bias moved with the participant, shifting to a previously sparse quadrant on the screen. Incidental learning of a target's likely locations led to a persistent, egocentric spatial bias. (Contains 1 table and 7 figures.)