Abstract
A filtering paradigm was used to study selection by feature-based attention in transparent motion displays. Observers viewed a single field of dynamic random dots with net motion in several possible directions. The task was to discriminate between relevant directions of motion while ignoring irrelevant directions of motion. For example, one might discriminate between leftward and rightward motion while ignoring diagonal motions. By manipulating the motion strength in both relevant and irrelevant directions, one can test for selection by blocking versus selection by attenuation. With blocking, withdrawing attention prevents detection of even a strong stimulus; with attenuation, withdrawing attention can be overcome by a sufficiently strong stimulus. The results were consistent with blocking and not attenuation. They rule out models of attenuation such as a motion analog to the contrast gain model of contrast detection. Possible models of blocking include attention switching, response gain, or a selection process in decision rather than perception.