Abstract: A study was conducted to determine the relationship between discrimination reversal learning and performance tasks in horses. Ten yearling and seven 2-year-old mares and geldings of Arabian (n = 4), Quarter Horse (n = 9), and Thoroughbred (n = 4) breeding were given a two-choice discrimination task in which either a black or a white bucket contained a food reward for ten trials per day during 19 test days. The spatial position of the buckets was varied on a random schedule. The rewarded bucket color was reversed each time a subject met criterion of eight correct choices per day for 2 consecutive days. Discrimination reversal testing was followed by 6 days of performance tasks: three crossing a wooden bridge and three jumping an obstacle to reach food and conspecifics, within a maximum allotted time of 15 min day-1. Total reversals attained by the horses were low (x = 1.5 +/- 0.9). All subjects did attain at least one reversal, and six had two or more reversals. No differences (P > .05) were detected between ages or sexes, nor among breeds in discrimination reversal learning or performance test measurements. However, there was a trend towards a breed difference (P <= 0.09) in the mean number of correct responses to the first reversal criterion. Correlations between reversal learning results and performance task results were extremely low, indicating that the discrimination reversal learning test was not useful for predicting success at these performance tasks. Results from the two performance tasks also showed little correlation (r = 0.04, P < 0.91), indicating that horses might not use the same approach when solving the problem of crossing these two obstacles. The overall poor performance of the horses on the discrimination reversal task suggests horses may have difficulty reversing previously learned tasks.