Abstract: Fourteen riding horses of mixed age and breed were randomly allocated to observer and control treatments. An additional horse was pre-trained as a demonstrator to walk the 13.8 m length of the test arena and select one of two food buckets using colour and pattern cues. Observer horses were exposed to correct performances of the task by the trained demonstrator, for 20 trials held over 2 days. Control horses were subjected to the same handling and placement procedures as the observer horses but without exposure to the behaviour of the demonstrator. The third day for all subjects was designated as a test day. Each subject was released individually in a predetermined place in the arena, and the latency to walk the length of the test arena to the food buckets, the latency to feed, the identity of the bucket approached and the identity of the bucket selected were recorded on ten consecutive trials. During tests both food buckets contained food to minimize the possibility of individual trial and error learning. On the first trial the mean latency to approach the goal area was 18 s for observer horses, compared with 119 s for control horses (t = 2.8, d.f. = 12, P < 0.01) and the mean latency to eat was 35 s for observer horses, compared with 181 s for control horses (t = 4.86, d.f. = 11, P < 0.001). However, observer horses were no more likely to choose the demonstrated bucket than control horses on the first trial. Twelve of the 14 horses decreased their latency to approach the goal area during the series of ten trials, but there were no significant changes in the buckets selected.