Twenty-four participants learned to choose one of two buttons to obtain a reward. On each trial, one of two shapes appeared on a screen while one of two tones was played. There were four shape–tone combinations, and the correct button depended on the combination: neither the shape alone nor the tone alone predicted the correct button. Participants received feedback during four training blocks. Twenty-four hours later, they completed a test without feedback or further practice. With two possible buttons, chance performance is 50%.

The graph shows the mean percentage of correct choices in each block and at the 24-hour test. Means are rounded to the nearest whole percentage point.