Inhibition of return (IOR) refers to a performance cost associated with previously attended locations (or items). It has been suggested that this inhibitory attention mechanism evolved to maximize visual sampling (Posner & Cohen, 1984) and can facilitate visual search by decreasing the probability of attention returning to previously inspected locations (Klein & MacInnes, 1999). The reference frame of IOR is one of the hotly debated theoretical questions in the IOR literature. Previous studies have shown that IOR can be encoded in retinotopic, spatiotopic, and even object-centered representations. In Posner and Cohen's (1984) seminal paper on IOR, it was reported that IOR resided in spatiotopic representations. By improving the cueing task Posner and Cohen (1984) used to probe spatiotopic IOR, Maylor and Hockey (1985) confirmed Posner and Cohen's (1984) conclusion, showing that only spatiotopic IOR could be revealled when the spatiotopic and retinotopic coordinates were dissociated with a saccade intervening the cue and target. More recent studies, however, have shown that IOR can be coded in both spatiotopic and a retinotopic presentations (Hilchey, Klein, Satel, & Wang, 2012; Math?t & Theeuwes, 2010; Pertzov, Zohary, & Avidan, 2010; Satel, Wang, Hilchey, & Klein, 2012). With a graphical meta-analysis of the data available in the literature, we show that the spatiotopic IOR is relatively stable over time, whereas the retinotopic IOR quickly dissipates following the intervening saccade. In addition to the stability of the visual environment, there is a pressing need to keep track of objects in the dynamic world as stable, continuous entities. Tipper, Driver, and Weaver (1991) found that IOR could be tagged to a placeholder after it moved to a new spatial location, suggesting that IOR can also be coded on object-centered representations. Tipper et al.'s (1991) finding was confirmed and extended in several follow-up studies (e.g., Tas, Dodd, & Hollingworth, 2012; Tipper, Jordan, & Weaver, 1999; Tipper, Weaver, Jerreat, & Burak, 1994; Weaver, Lupiafiez, & Watson, 1998; but see Mu?ller & von Mu?hlenen, 1996). These studies have shown that a) the object-based IOR can reside not only in dynamic environment but also in static scenes, b) the object-based IOR coexists with spatitopic IOR and these two forms of IOR have additive behavioral effect, c) the activation of the oculomotor system in a cueing task will evoke both the sensory and motor components of IOR, however, only the sensory component can be coded on object-centered coordinates (Abrams & Dobkin, 1994), and d) IOR can also reside in the relative space defined by an object (Gibson & Egeth, 1994). Consistent with the findings of previous lesion studies, a recent EEG study showed that retinotopic and spatiotopic IORs were associated with a weakening of the P1 and Nd event-related potential components, respectively, at parieto-occipital electrode sites (Satel et al., 2012). The neural basis of object-centered IOR is less clear, with studies suggesting that it is likely coded in cortical regions that are more susceptible to aging (McCrae & Abrams, 2001). In addition to an overview of the experimental paradigms used to explore the reference frame of IOR and related behavioral, imaging an lesion findings, issues that should be addressed in future studies were also discussed in the end.