Healthcare practitioners such as physicians or nurses often underestimate patients' well-being impairment (e.g., pain, anxiety) which may lead to undesirable consequences on treatment decisions. Lack of recognition/identification of signals and over-exposure are two reasons invoked, but a combination of factors may be involved. Studying human decoding of animals' expressions of emotions showed that "identification" to the subject was necessary to decode the other's internal state. In the present study we wanted to compare caretakers' reports on the prevalence of stereotypic or abnormal repetitive behaviors, to ethological observations performed by an experienced observer on the same horses in order to test the impact of these different factors. On the first hand, a questionnaire was given hand to hand to the caretakers. On the other hand, the experienced observer spent 18 h observing the horses in each stable. Here we show that caretakers strongly underestimate horses' expressions of well-being impairment. The caretakers who had a strong concern about their horses' well-being were also those who reported the more accurately SB/ARB's prevalence, showing that "identification" to the subject is a primary factor of bad-being signal's detection. Over-exposure also appeared to be involved as no SB/ARB was reported in stables where most of the horses were performing these abnormal behaviors. Being surrounded by a large population of individuals expressing clear signals of bad-being may change professionals' perceptions of what are behaviors or expressions of well being. These findings are of primary importance as (1) they illustrate the interest of using human-animal relationships to evaluate humans' abilities to decode others' states; (2) they put limitations on questionnaire-based studies of welfare.