In studies using Averbach and Coriell's (1961) partial-report bar-probe paradigm with linear arrays, most errors involve the naming of an item that was in the display but in a position other than the cued one. Up to now, there is no general agreement on the origin of these location errors. Point of departure in this paper is that part of the location errors arises from inappropriate application of the cue. It is tested whether this originates from problems to perceive the position of the cue ("cuedisplacement hypothesis") or from confusion about the order of the items in the array ("item-order hypothesis"). The results of two bar-probe experiments are reported. A novel, crucial, finding in both experiments is that, among the location errors, there was a preponderance of response letters that came from the central side of the cued item. In the second experiment, this was observed not only in the usual postcue conditions but also when the cue preceded the array. These results positively corroborate the cue-displacement hypothesis and do not support the item-order hypothesis: The cue tends to be perceived more toward the center of the visual field than it actually is exposed-that is, there is a central drift of the cue.The partial-report technique is a powerful tool in the study of selective attention and memory processes in vision. In this technique, subjects are briefly exposed to a configuration of letters and a cue that specifies which part of the display is to be reported. In the variety introduced by Averbach and Coriell (1961 )-the partialreport bar-probe paradigm-one or several rows of letters are displayed, and only one letter has to be named. A visual cue (e.g., a bar or an arrow) is used to indicate the target letter. The cue is presented just before, during, or at various intervals after exposure of the letter display.From these partial-report bar-probe studies, a consistent pattern of results has emerged (see, e.g., Averbach & Coriell, 1961;Hagenaar & van der Heijden, 1995;Mewhort, Campbell, Marchetti, & Campbell, 1981;Mewhort, Marchetti, & Campbell, 1982;Townsend, 1973). The first robust finding is that accuracy of report depends on the moment of appearance of the cue. If the cue just precedes the letter display or is shown simultaneously, accuracy is quite high (approximately 75% correct). If the cue follows the display immediately, accuracy of report is still high. Accuracy decreases gradually when the cue is further delayed. At a delay of about 250 msec after display termination, accuracy reaches an asymptotic level ofabout 35% correct. The second robust finding is that accuracy is higher for letters displayed at the center and the ends of the rowes) than for letters shown between the center and Correspondence should be addressed to A. H. C. van der Heijden, Unit of Experimental and Theoretical Psychology, Leiden University, P.O. Box 9555, 2300 RB Leiden, The Netherlands (e-mail: heijden@ rulfsw.leidenuniv.nl ).end positions: The percentage correct reports are distributed in the shape of a "W" acro...