The perception-action model proposes that vision for perception and vision for action are subserved by two separate cortical systems, the ventral and dorsal streams, respectively [Milner, A. D., & Goodale, M. A. (1995). The visual brain in action (1st ed.). Oxford: Oxford University Press; Milner, A. D., & Goodale, M. A. (2006). The visual brain in action (2nd ed.). Oxford: Oxford University Press Inc.]. The dorsal stream codes spatial information egocentrically, that is, relative to the observer. Egocentric representations are argued to be highly transient; therefore, it might be expected that egocentric information cannot be used for spatial memory tasks, even when the visual information only needs to be retained for a few seconds. Here, by applying a spatial priming paradigm to a visual search task, we investigated whether short-term spatial memory can use egocentric information. Spatial priming manifests itself in speeded detection times for a target when that target appears in the same location it previously appeared in. Target locations can be defined in either egocentric or allocentric (i.e. relative to other items in the display) frames of reference; however, it is unclear which of these are used in spatial priming, or if both are. Our results show that both allocentric and egocentric cues were used in spatial priming, and that egocentric cues were in fact more effective than allocentric cues for short-term priming. We conclude that egocentric information can persist for several seconds; a conclusion which is at odds with the assumption of the perception-action model that egocentric representations are highly transient.