In order to understand visual attention, it is important to know whether we are able to limit selection to a goaldefined feature of the environment or whether selection is determined in a more bottom-up manner. Theeuwes (1991) examined the effectiveness of top-down selection by presenting two feature singletons in a visual search task. For example, in his Experiment 2, colored elements were arranged around a fixation point. Targets were defined by a unique attribute: a color or a shape. These singleton targets could also appear with singleton distractors, where the unique attribute was not defined as a target feature for the task. On trials in which the target was defined by its color, a distractor might have a unique shape. On trials in which the target was defined by its shape, the distractor might have a unique color. Subjects had to discriminate orientation of a line segment presented in a target. For targets defined by color, there was no difference in reaction times (RTs) when (shape) singleton distractors were or were not present. However, for targets defined by shape, RTs were slowed when the (color) singleton distractor was present, relative to when it was absent: There was cross-dimensional interference. Asymmetric cross-dimensional interference has subsequentlybeen reported with other dimensions, including color and luminance (Theeuwes, 1991), color and onset (Theeuwes, 1994), and color and orientation (Kumada, 1999). Cross-dimensional interference suggests that selection cannot be constrained solely by top-down factors, since stimuli from a dimension not defined for selection disrupt performance.The most important factor affecting asymmetric crossdimensional interference is the relative saliency of the target and the singleton distractor. Theeuwes (1991) showed that when the discriminabilityof color singletonsdecreased, shape singleton distractors interfered with search for a color target, whereas a color singleton distractor did not interfere with search for a shape target. To explain these results, Theeuwes (1991) argued that attention is captured by the most salient features in a display, irrespective of their relation to the goal of the task. When the most salient feature is the target, RTs are unaffected by the presence of an irrelevant singleton distractor. On the other hand, when the target is less salient than a singleton distractor, attention is unintentionallydrawn to the distractor, slowing RTs. Theeuwes (1991) concludedthat asymmetrical cross-dimensional interference provided evidence for a failure of top-down control over visual selection.Contrasting data have been provided by Folk and colleagues, using a form of spatial precuing (Folk & Remington, 1998;Folk, Remington, & Johnston, 1992;Folk, Remington, & Wright, 1994). In Folk and Remington, a target display consisted of four characters in four boxes around a fixation point. Subjects had to determine whether a character indicated by a target color was an "X" or an "5." Before target displays were presented, however, a distractor display...