How do bees employ multiple visual cues for homing? They could either combine the available cues using a view-based computational mechanism or pick one cue. We tested these strategies by training honeybees, Apis mellifera carnica, and bumblebees, Bombus terrestris, to locate food in one of the four corners of a box-shaped flight arena, providing multiple and also ambiguous cues. In tests, bees confused the diagonally opposite corners, which looked the same from the inside of the box owing to its rectangular shape and because these corners carried the same local colour cues. These ‘rotational errors’ indicate that the bees did not use compass information inferred from the geomagnetic field under our experimental conditions. When we then swapped cues between corners, bees preferred corners that had local cues similar to the trained corner, even when the geometric relations were incorrect. Apparently, they relied on views, a finding that we corroborated by computer simulations in which we assumed that bees try to match a memorized view of the goal location with the current view when they return to the box. However, when extra visual cues outside the box were provided, bees were able to resolve the ambiguity and locate the correct corner. We show that this performance cannot be explained by view matching from inside the box. Indeed, the bees adapted their behaviour and actively acquired information by leaving the arena and flying towards the cues outside the box. From there they re-entered the arena at the correct corner, now ignoring local cues that previously dominated their choices. All individuals of both species came up with this new behavioural strategy for solving the problem provided by the local ambiguity within the box. Thus both species seemed to be solving the ambiguous task by using their route memory, which is always available during their natural foraging behaviour.