According to the contingent capture hypothesis, observers can specify their control settings in advance of the target's presentation to quickly attend to relevant target colors. Two predictions were derived from this hypothesis and tested in a manual choice response task. First, contingent capture by color was expected: capture of spatial attention by a better-matching color stimulus should be stronger than capture by a less-matching color stimulus. Second, with the control settings specified in advance, the contingent capture by color should commence early after the stimulus onset and should be evident among fast correct responses in an RT distribution. Both predictions are shown to hold true in two experiments. Results are discussed in light of contrasting evidence for saccadic instead of manual responses.