Animals with visual perspective taking abilities should differ in their responses to individuals that do and individuals that do not have visual access to some critical event. We investigated whether domestic pigs show behaviour consistent with this ability. Ten subjects were trained to move from a start box into one of four corridors that they had seen a human enter with a bucket for baiting. They received a food reward for choosing the correct corridor. In unrewarded probe tests, the subjects' view of the corridors was blocked, but they could see a 'seeing' companion pig who had visual access to the baiting event, and another, whose view was also blocked, located in start boxes to their left and right. After the companions had been released and entered corridors, the subject was released and which companion it followed was recorded. Eight pigs followed companions less frequently than expected by chance, probably due to specific corridor or centre/side preferences. However, one subject showed no positional bias, and a significant preference for following the 'seeing' companion, consistent with the ability to take another's visual perspective. Our design rules out learning of a correct response during the experiment, by testing subjects in unrewarded probe trials, and by using companions that were equally trained and unaware of the subject's task. However, we cannot rule out the possibility that previously learned contingencies were used to solve the problem. We therefore consider that this apparent visual perspective taking ability does not necessarily imply any ability to understand knowledge states.