Moods can influence our judgment of ambiguous stimuli as positive or negative. Measuring judgment bias in animals is a promising method to objectively assess their emotional states. Our study aimed to develop a cognitive bias test in horses, in order to assess the effect of training using positive reinforcement (PR) or negative reinforcement (NR) on their emotional states. We trained 12 mares to discriminate between a rewarded and a non-rewarded location situated on each side of a paddock. The mares were then trained during five days to perform several exercises using PR (n = 6) for one group, and NR (n = 6) for the other (treatment). Finally, we compared the responses of the two groups to three ambiguous locations situated between the rewarded and non-rewarded locations (judgment bias test). During the training exercises, according to our predictions, behavioural measures suggested that NR mares experienced more negative emotions than PR mares. Surprisingly, the results of the judgment bias test suggest that NR mares were in a more optimistic mood compared to PR mares, despite previously experiencing more negative emotions during the treatment. NR mares could have been more motivated to obtain a food reward than PR mares, which had been rewarded throughout the treatment phase. Alternatively, NR mares could have developed optimistic bias triggered by release from the negative state experienced during treatment. This first attempt to test judgment bias in horses suggests that this is a promising method to measure horse mood. Knowledge about the effect of training methods on the mental health of domesticated animals can add a new dimension to animal welfare, in order to promote better ways to work with animals.