Self-referential stimuli such as self-face surpass other-referential stimuli in capture of attention, which has been attributed to attractive perceptual features of self-referential stimuli. We investigated whether temporarily established self-referential stimuli are different from other-referential cues in guiding voluntary visual attention. Temporarily established self-referential or friend-referential shapes served as central cues in Posner's endogenous cueing task. We found that, relative to friend-referential cues, self-referential cues induced smaller cueing effect (i.e., the difference in reaction times to targets at cued and uncued locations) when the interstimulus interval was short but larger cueing effect when the interstimulus interval was long. Our findings suggest that temporarily established self-referential cues are more efficient to capture reflexive attention at the early stage of perceptual processing and to shift voluntary attention at the later stage of perceptual processing.