Two groups of Clark's nutcrackers were trained to find buried seeds whose l
ocation was defined by a constant angle from two landmarks whose interlandm
ark distance and position in the room varied across trials. The first group
had a landmark array that was always placed in the same orientation with r
espect to the walls, allowing the animals to use both relative and absolute
bearings. The second group had a landmark array that rotated across trials
so that only relative bearings could be used to locate the seeds. The bird
s in each group learned the task and transferred to new interlandmark dista
nces both within and beyond the range of training distances. Results from t
hese experiments indicate that nutcrackers can learn a geometric relationsh
ip that relies exclusively on relative bearings even though the use of abso
lute bearing yields more efficient search.