In order to solve the aperture problem, and to reduce noise, motion estimates are integrated over space. Nishida et al (2006, Journal of Vision, 6(6):1084) showed that arrays of randomly oriented 1D motion elements appear to move coherently in a single direction if the speeds are consistent with a specific 2D motion. It is generally assumed that motion integration fields are radially symmetric. If so then coherence should be unaffected by rotating the global configuration of the motion elements with the distribution of local velocities held constant. To test this we compared the apparent speeds of arrays of Gabor and Plaid elements (1 or 2 superimposed sine gratings modulated by Gaussian windows). Subjects were shown arrays of Plaids drifting in one direction followed by spatially identical arrays of randomly oriented Gabors drifting in the same global direction and were asked to judge which was moving faster. The 50% point on the psychometric function was used as a measure of the perceived speed of the Gabors. At high densities the two stimuli were comparable, but the relative perceived speed of the Gabors fell as the number of patches decreased. The reduction in perceived speed can be interpreted as resulting from a loss of coherence accompanied by a shift from an intersection of constraints calculation towards a vector average speed estimate. We also found that Gabors arranged in a line appeared to move slower than when arranged randomly. We then compared global motion parallel or orthogonal to the in line spatial layout. Global motion appeared slower when orthogonal, rather than parallel, to the line. This indicates that the integration zone underlying the global motion computation must be dynamically elongated in the direction of motion. This asymmetry is reminiscent of the 'association fields' of contour and motion integration.