Abstract: Twenty-six yearling horses were tested on a serial reversal learning discrimination combining spatial and brightness cues. An original discrimination of rewarded or nonrewarded stimuli was made followed by 20 daily reversals for position/brightness discrimination. Learning criteria were defined as 11 out of 12 correct, with the last eight responses all correct. Each horse was allowed 30 trials per discrimination to achieve criteria. Mean errors (ME) and mean trials (MT) required to achieve criteria were computed for each horse. A relative learning ability index (LAI) was calculated by the formula 1000/MT/ME. A daily emotionality score, based on a scale of one (tranquil) to six (very excitable) was assigned each horse each day after testing and a mean computed for each horse. A single subjective trainability score, based on a scale of one (difficult to train) to six (easy to train) was obtained for each horse from an independent trainer. Linear regression analyses for all 26 horses revealed a reduction in MT and ME (P<.01) over the 21-day test period indicating evidence of learning to learn. Differences (P<.05) were evident between sexes for MT and ME. Significant correlations between trainability scores and learning ability indices MT, ME, and LAI were evident for colts and geldings but not for fillies. Pooled data showed significant correlations between ME and trainability. There was a negative correlation (P<.05) between emotionality and trainability scores for all 26 horses, although the filly group did not exhibit significant correlation between these parameters.