We previously proposed a model for detecting local image velocity on the magnocellular visual pathway (Kawakami & Okamoto (1996) Vision Research, 36, 117-147). The model detects visual motion in two stages using the hierarchical network that includes component and pattern cells in area MT. To validate the model, we predicted two types of bimodal direction tuning for MT neurons. The first type is characteristic of component cells. The tuning is bimodal when stimulated with high-speed spots, but unimodal for low-speed spots or for bars. The interval between the two peaks widens as the spot's speed increases. The second type is characteristic of pattern cells. The tuning is bimodal when stimulated with low-speed bars, but unimodal for high-speed bars or for spots. The interval widens as the bar's speed decreases. To confirm this prediction, we studied the change of direction tuning curves for moving spots and bars in area MT of macaque monkeys. Out of 35 neurons measured at various speeds, six component cells and four pattern cells revealed the predicted bimodal tunings. This result provided neurophysiological support for the validity of the model. We believe ours is the first systematic study that records the two types of bimodality in MT neurons.