Individually foraging desert ants of central Australia, Melophorus bagoti, exhibit amazingly precise mechanisms of visual landmark guidance when navigating through cluttered environments. If trained to shuttle back and forth between the nest and a feeder, they establish habitual outbound and inbound routes, which guide them idiosyncratically across the natural maze of extended arrays of grass tussocks covering their foraging areas. The route-based memories that usually differ between outbound and inbound runs are acquired already during the first runs to the nest and feeder. If the ants are displaced sideways of their habitual routes, they can enter their stereotyped routes at any place and then follow these routes with the same accuracy as if they had started at the usual point of departure. Furthermore, the accuracy of maintaining a route does not depend on whether homebound ants have been captured at the feeder shortly before starting their home run and, hence, with their home vector still fully available (full-vector ants), or whether they have been captured at the nest after they had already completed their home run (zero-vector ants). Hence, individual landmark memories can be retrieved independently of the state of the path-integration vector with which they have been associated during the acquisition phase of learning. However, the ants display their path-integration vector when displaced from the feeder to unfamiliar territory.