||Humans have domesticated many kinds of animals in their history. Dogs and horses have particularly close relationships with humans as cooperative partners. However, fewer scientific studies have been conducted on cognition in horses compared to dogs. Studies have shown that horses cross-modally distinguish human facial expressions and recognize familiar people, which suggests that they also cross-modally distinguish human emotions. In the present study, we used the expectancy violation method to investigate whether horses cross-modally perceive human emotions. Horses were shown a picture of a human facial expression on a screen, and they then heard a human voice from the speaker before the screen. The emotional values of the visual and auditory stimuli were the same in the congruent condition and different in the incongruent condition. Horses looked at the speaker significantly longer in the incongruent condition than in the congruent condition when they heard their caretaker's voices but not when they heard the stranger voice. In addition, they responded significantly more quickly to the voice in the incongruent condition than in the congruent one. To the best of our knowledge, this is the first study to show that horses cross-modally recognized the emotional states of their caretakers and strangers.