The present study demonstrated that the magnitude of after-effect due to wedge prisms depends on the form of the visual feedback used to represent hand and target position in fast, targeted, transverse reaches. Trained human subjects made reaches with and without prisms in three visuomotor representations (VR): (1) the subject's actual hand and targets (Direct), (2) a real-time video broadcast of hand and targets (Video), or (3) abstract, computer-generated targets and a cursor representing hand position (Cursor). A significant after-effect occurred in each VR. However, the magnitude of the after-effect was significantly different among VRs: the magnitude was greatest in Direct, smaller in Video and smallest in Cursor. A significant after-effect (carryover) also occurred when a subject prism-adapted reaches in one VR and then removed the prisms and made initial reaches in another VR. Our data showed that when reaches were prism-adapted in Direct and then prisms were removed, there was a large carryover to initial reaches in Video or Cursor (D-->V and D-->C). In contrast, when prisms were worn in Video and removed for reaches in Direct (V-->D), there was a significantly smaller carryover than from both D-->V and D-->C. Finally, when prisms were worn in Cursor and removed for reaches in Direct (C-->D), there was very little detectable carryover. Our results suggest that adaptation is context-dependent and that the magnitude of carryover is dependent on the VR in which adaptation occurred. Interpretations of adaptations made in abstract training and experimental conditions may be greatly affected by this finding.