Path integration was tested in golden hamsters, Mesocricetus auratus, within a new environment and in darkness. From the exit of its home cage, the subject was lured over a certain distance into the experimental arena, and then startled by a sudden noise to induce instantaneous homing. The subjects in experiment 1 had previously been tested within an identical arena in another building; they started the outward journey from a peripheral nest. Upon hearing the sound, all subjects returned in the correct direction to the arena border. In experiments 2 and 3, experimentally naïve subjects started the outward trip from a central nest exit. Both groups initially oriented correctly, but only a few of subjects returned to the nest. According to computational considerations, experiment 1 required mainly the calculation of the homing direction, while experiments 2 and 3 required a homing vector. Furthermore, the experienced subjects may have been helped by their homing experience in the same type of arena, while the naïve subjects may not have ‘trusted' self-generated vector information sufficiently to follow it to the goal in an uncharted environment. Thus, the performance, but not the computation of self-generated vector information, may depend on the animal's experience.