Children between 1.5 and 4 years old were tested for their ability to relocate a hidden object after a 180° self-produced movement around an array of four locations. In one task the object′s location relative to the other locations could be uniquely defined within one dimension, while in another two dimensions were needed to do this. No differences emerged between conditions, and by 3 years few errors occurred, despite the fact that children were unable to view the array during movement. This indicates either that young children encounter no specific difficulty in coordinating dimensions or that they solved the task without recourse to such a system. An error analysis supports the second possibility. Children apparently tackled the task by a system directly related to body movement, since errors were frequently the result of incomplete compensation for movement around the array. In a second study in which the four containers were placed in contact, children′s performance declined and the relation between direction of movement and error was replaced by some evidence for updating on the near-far dimension accompanied by failure to update the left-right dimension. Thus children appear to change strategy when the problem requires more precise specification of target location.