Nine macaque monkeys were trained to discriminate between a display in which all the dots moved randomly (visual dynamic noise) and an adjacent display in which a proportion of the dots oscillated coherently from side to side. Two monkeys in which area MT (middle temporal visual area) and adjacent regions were surgically removed were unable to perform even the simplest version of this task where the coherent motion was not masked by any visual dynamic noise. The other seven animals tolerated between 60% and 65% of random movement in the otherwise coherently oscillating display before they failed to discriminate between the two displays. In contrast, shape discrimination tasks containing luminance boundaries were unaffected by the removal of area MT. However, the removal of area MT did impair the animals' ability to use a kinetic boundary as the basis for shape discrimination. Their performance with a kinetic boundary between a moving and a stationary random dot field and a kinetic boundary between two random dot fields moving in opposite directions was the most severely impaired. Their performance with a kinetic boundary defined by two random dot fields moving orthogonally to each other was unimpaired. None of the animals from either group was able to use the kinetic boundary defined by coherent motion and visual dynamic noise as the basis for shape discrimination. The ease with which the control animals were able to employ the different types of kinetic boundaries to perform the shape discrimination reflected previously published proportions of cells in area MT that are specialized for detecting the different types of kinetic boundaries.