The relative effectiveness of central arrow, peripheral arrow, and abrupt-onset cues was assessed in a character recognition task. On each trial, either a central or a peripheral arrow cue was presented 0, 100, or 200 msec before the appearance of a three-digit display. Two of the digits were "uncamouflaged" from previous figure-eight masks, whereas the third digit appeared abruptly in a previously empty space. Four different groups of subjects were run in factorial combinations of high or low expected validities for arrow and onset cues. In Experiment 1, arrow cues were located centrally, near the fixation point. Abrupt onsets showed larger cost-plus benefits than central arrows, except when subjects expected the central cues to have higher validity than the onsets. In Experiment 2, arrow cues were located peripherally, near the display digits, and abrupt onsets were again more effective in capturing attention except when peripheral cues had higher validity and led the onsets by 100 msec or more. In both experiments, the relative effectiveness of abrupt onsets decreased with arrow SOA. The results were consistent with a model in which automatic and voluntary processes interact in their control of attentional resources.