• Purpose.
  If the elements of a random dot cinematogram (RDC) move according to not one direction but some distribution of directions, the result will be a percept of intermingled local movements. If the distribution of directions is sufficiently narrow, another percept is produced: At the same time that the subject sees the local, intermingled movements, the subject also perceives global flow approximately along the cinematogram's mean direction. To un derstand attention's role in these percepts, we studied responses to RDCs that were composed of two separate sets of dots. Each set moved according to its own distribution of directions. We asked whether subjects could deploy attention to (a) determine th e mean direction of global flow for both sets of dots together, and (b) determine the mean direction of flow of just one set of dots, attempting to ignore the other set of directions. We also asked whether a contrast difference between the two sets of dots would facilitate subject 's attention to just one set of directions.

• Methods.
  Each 80 ms RDC showed six frames of black and/or white dots on a grey background. From frame to frame of the cinematogram, the dots took a constrained random walk of uniform step size, 5.6 arcmin. In the Narrow and Wide conditions, all dots were black and drew their step directions from distributions that spanned 30 degrees and 135 degrees, respectively. In the Bimodal condition, the directions of two sets of black dots were drawn from two distributions, each spanning 30 deg, and separated from one another by a gap of 75 deg. The Polarized condition was the same as the Bimodal, except that one set of dots was white and one was black. Subjects estimated the global mean direction of each type of RDC, and on separate trials also estimated the mean direction of the most clockwise set of Bimodal and Polarized RDCs.

• Results.
  Error was defined as the mean absolute direction error across trials. In estimating direction of global flow, subjects' asymptotic error on Bimodal RDCs was comparable to that on Wide RDCs, even though in the former stimulus no dots actually moved along the global mean direction, the direction subjects were estimating. In estimating direction of clockwise flow in Bimodal RDCs, error was far higher than in estimating global flow in Narrow RDCs, indicating that subjects were unable to ignore the counterclockwise population. Preliminary results indicate that error in estimating clockwise flow in Polarized RDCs is no lower than in Bimodal RDCs, suggesting either that attention cannot be restricted to a single-polarity population in this task, or that other effects of the polarity difference overwhelm any such benefit.

• Conclusions.
  Subjects can integrate motion vectors over a wide range of directions, but are poor at attending to a narrow subset of directions. Preliminary results indicate that subjects are also unable to attend to a single dot polarity in this task. We are currently investigating whether there are similar tasks in which subjects can attend to only one polarity, and whether subjects can integrate directions across polarity.

Support: NSERC & James S. McDonnell Foundation.