ABSTRACTJudgment bias tasks for nonhuman animals are promising tools to assess emotional valence as a measure of animal welfare. In view of establishing a valid judgment bias task for horses, the present study aimed to evaluate 2 versions (go/no-go and active choice) of an auditory judgment bias task for horses in terms of acquisition learning and discrimination of ambiguous cues. Five mares and 5 stallions were randomly assigned to the 2 designs and trained for 10 trials per day to acquire different operant responses to a low-frequency tone and a high-frequency tone, respectively. Following acquisition learning, horses were tested on 4 days with 3 ambiguous-tone trials interspersed between the 10 high-tone and low-tone trials. All 5 go/no-go horses but only one active-choice horse successfully learned their task, indicating that it is more difficult to train horses on an active choice task than on a go/no-go task. During testing, however, go/no-go horses did not differentiate between the 3 different ambiguous cues, thereby making the validity of the test results questionable in terms of emotional valence.