The aim of the study was to determine whether adult horses can learn to recognise a novel word and how important the familiarity of a human’s voice uttering the taught word is. In the study, 18 adult warmblood horses were the subjects: nine mares and nine geldings. At the beginning, the horses were habituated to voices heard from loudspeakers located near boxes for three days. The horses were listening to two familiar people (A and B) and one unfamiliar person (C) whose voices were recorded in the playback individually. Next, each horse was taught to positively associate a novel word with receiving a titbit for 10 days, 15 times daily: 12 times by the person who uttered the word and 3 times when the word was heard from playback. The same three people were engaged during the teaching. Finally, the horses were tested whether they had learned the word. The word was uttered by one familiar person (A), the initially unfamiliar person (C) and a totally unknown person (D). The voices of these people were played solely from the loudspeaker three times daily. The horses’ response was evaluated by nine kinds of behaviour recorded. The first test with the three people’s voices was repeated for up to five three-day cycles, dependent on manifesting by the horses a minimum number of behaviours per day. Apart from each tested horse, one companion horse remaining in a neighbouring box and taught another word with a titbit was studied to compare the response. The results show that horses perceived differences between single words, and associated a taught word with receiving a titbit. The companion horses did not respond to this word. The responses to familiar or unfamiliar people’s voices were similar. It was concluded that horses are capable of discriminating a taught word and, simultaneously, generalise familiar and unfamiliar human voices uttering the word. These findings show that teaching horses single words is effective. Simultaneously, the results highlight the need to consistently use chosen words when teaching horses.