Abstract: This experiment was designed to determine if a horse could learn the location of grain by watching another horse find grain in one of two feed buckets. Both experimental and control groups contained 9 quarter horses consisting of five 2-year-old mares, two 2-year-old geldings, and two 3-year-old geldings. Two mature geldings were used as “demonstrators”. An “experimental” was a horse that could watch three times daily another horse, the “demonstrator”, choose between and eat grain from a black or white bucket, only one of which contained grain. A “control” was a horse that could watch a demonstrator in the same arena for 3 min daily when both feed buckets were removed. When the demonstrator was removed on each of 15 successive days, the experimental or control horse was given five trials to determine if it could find the feed bucket with grain. No significant difference between experimentals and controls occurred for both first and total correct choices and for time to reach the feed bucket with grain. We conclude that no observational learning occurred. This experiment was also used to determine if the identity of horses that learned rapidly by trial and error could be predicted by the time it took to reach the feed bucket with grain. Data from the last three trials of experimentals and controls were combined. Significantly less time to find feed was needed by horses with more than the median number of correct choices. Both number of correct choices and time needed to contact a feed bucket summed over the first 5 days accurately predicted the same data summed over the last 10 days. We conclude that horses that learn rapidly by trial and error make correct choices rapidly, and that these horses can by identified after 5 days of testing.