Rats were trained to select a final, remaining baited arm following a 6- to 10-min delay following their entries into three experimenter-selected baited arms in an enclosed 4-arm radial maze containing different proximally cued arms. Rats' accuracy in selecting the remaining baited arm was disrupted when the spatial configuration of arm cues was randomly varied over trials following initial training with one configuration in Experiment 1. In Experiment 2, the same rats acquired this task with the original and a new configuration of the same arm cues when each consistently occurred at a specific time of day (one in the morning, the other in the afternoon). Randomly varying the temporal presentations of these configurations following acquisition disrupted rats' choice accuracy more within the new than the original configuration. Other rats in Experiment 3 learned this task with two configurations containing different types of arm cues (full arm inserts, objects at the arm entrances). When required to relearn this task with recombined configurations of pairs of arm cues from of each configuration, only rats presented pairs of arms arranged differently from that in their original configurations were unable to reacquire the task. Together these results support a cognitive map hypothesis more than a proximal arm cue list hypothesis. These findings were discussed in terms of recent versions of cognitive map theory (Benhamou, 1998; Poucet, 1993) and the possible limits of such processing (Roberts, 2001).