We tested normal observers with a motion discrimination task in which signal and noise dots were presented with different binocular disparities giving the appearance of two planes in depth. By varying the depth separation between the two planes, we found that observers were more impaired in the presence of near-disparity noise than far-disparity noise. Two control experiments suggested that the near-far differences could not be accounted for by attention or surface completion processes. Therefore, we sought an explanation at the physiological hardware level, and developed a physiologically constrained model of neurons in the middle temporal area (MT)—a cortical area known to significantly contribute to motion processing. Results of simulations with the model showed that when population anisotropies, like those reported by DeAngelis et al (2003), are included, the model predicts the near-far disparity skew we observed in the human psychophysical task. We will show that our psychophysical results can be explained by the model, and suggest that disparity tuning properties of MT are sufficient for producing depth segmentation of the motion signals.