Ego motion and natural motions in the world generate complex optic flows in the retina. These optic flows, if produced by rigid surface patches, can be decomposed into four components, including rotation and expansion. We showed previously that humans can precisely estimate parameters of these components, such as the angular velocity of a rotational motion and the rate of expansion of a radial motion. However, natural optic flows mostly display motions containing a combination of more than one of these components. Here, we report that when a pure motion (e.g., rotation) is combined with its orthogonal component (e.g., expansion), no bias is found in the estimate of the component parameters. This suggests that the visual system can decompose complex motions. However, this decomposition is such that the presence of the orthogonal component increases the discrimination threshold for the original component. We propose a model for how the brain decomposes the optic flow into its elementary components. The model accounts for how errors in the estimate of local-velocity vectors affect the decomposition, producing the increase of discrimination thresholds.