Rats were trained to find the hidden platform in a Morris pool, whose location was defined by reference to a small number of landmarks around the circumference of the pool. In each of three experiments, an experimental group was trained on alternate trials with two different subsets of three of the available landmarks, with the two subsets sharing one landmark in common. When tested with landmarks drawn from both of their training configurations, but without the landmark common to the two sets, they had no difficulty in locating the platform. In Experiment 1, they performed at least as well as a group trained with all the available landmarks present on every trial. In Experiment 2, they performed significantly better than a group trained with two different subsets of landmarks that shared no one landmark in common.