Bumblebees (Bombus terrestris) were trained to find one of the four exit holes located in the corners of an enclosed environment with a distinctive geometry (a rectangular cage). Panels located at the corners provided nongeometric, featural cues. Between trials bumblebees were passively disoriented to disable dead reckoning. When tested after removal of the panels, bumblebees reoriented using the residual information provided by the geometry of the cage. When tested after removal of only the two panels located in the two geometrically correct corners (the one with the exit and the diagonally opposite one), bumblebees were not able to use features in corners distant to the goal to reorient themselves. Finally, when geometric and featural cues provided contradictory information, bumblebees showed more reliance on featural cues. A similar outcome was observed when the conflict between geometrical and featural information was determined by first training bumblebees in a rectangular cage with a single wall of a different color used as a feature, and then testing animals with the feature displaced along a different wall. When the feature was close to the goal during training, bumblebees chose the corners with the feature at test, when the feature was far from goal during training, bumblebees chose the corners with the correct geometry at test. These results are similar to those revealed by similar transformational tests carried out in vertebrates relying mainly on vision for spatial orientation, that is, birds and monkeys.