Motion contrast describes a systematic pattern of error in apparent motion direction that occurs when two independently-moving groups of elements are transparently superimposed (Marshak and Sekuler, 1979). Dyre, Richman, and Fournier (2000) found a similar motion contrast effect on the localization of the foci of expansion in transparently superimposed diverging optical flow. Observers perceived a single coherent flow with an FOE between the actual FOEs for angular separations between the FOEs of less than 8° (attraction error), and two transparent flows with FOEs repulsed away from the actual FOEs (repulsion error) for greater angular separations. Similarly, Duffy and Wurtz (1993) showed that planar motion shifts the apparent location of the FOE of a diverging flow in the direction of planar motion. Here we examined the responses of linear sub-space heading models (Heeger & Jepson, 1992; Lappe & Rauchecker, 1995) to optical flow exhibiting non-rigid motion transparency to determine whether these models can account for these effects. Results show that the Heeger & Jepson (1992) model accurately predicts attraction errors, but is incapable of recognizing multiple radial flows. The 2-layer (emulating MT and MSTd) linear sub-space model described by Lappe & Rauschecker (1995) accurately predicts attraction errors and is capable of recognizing multiple flows, but does not predict repulsion errors, although it has accurately modeled the Duffy and Wurtz illusion (Lappe & Duffy, 1998). Lappe and Duffy theorize the illusion is an emergent property of the MST layer population response. We speculate that when two radial flows are presented symmetrically about the center of the field of view at angular separations less than 16° any bias in population response is nulled by an opposing bias. Current research is investigating whether a competitive layer of directional tuned MT neurons might explain the Dyre et al. (2000) results.