As stereotypic behaviours may reflect suboptimal life conditions, evaluating their prevalence in different environments is crucial in order to identify the problems and favour better management practices. In horses these surveys identified 1 to 10% of individuals in the tested populations performing stereotypic behaviours. However, a few observational studies converge towards much higher prevalence (22 to 96%). Different factors may explain this discrepancy, i.e. type of horse, definition of stereotypies or the methodology used. In order to disentangle these factors, we conducted a study on 373 horses of various ages and breeds, distributed amongst 26 riding schools. Questionnaires were given to caretakers, asking about the potential presence and type of stereotypies for each horse. In parallel, each horse was observed using Instantaneous Sampling method (3 sessions of 30 min). Horses were also observed continuously during 3 sessions of 6 h, to build an exhaustive list of stereotypic behaviours for each horse. The results showed a high discrepancy between the 2 types of evaluations both in the percentage of stereotypic horses detected (5.36% (Questionnaires) versus 37.53% (Observations), χ2-test P<0.001) and in the type of stereotypies detected (e.g. head nodding vs. weaving, χ2-test P<0.001). These results show that some stereotypies, even in ‘classical’ ones seem to be difficult to detect and abnormal repetitive behaviours (e.g. repetitive licking) may also not be identified as such. Thus, the under evaluation by caretakers, due to a lack of attention or recognition of stereotypies may be the primary factor in the differences observed between studies in the literature, but other aspects (unwillingness to admit the ‘problem’, horses having been punished when performing stereotypic behaviours, and performing less when in presence of caretakers...) remain to be tested. This study is raising the general question of using questionnaires in the assessment of behavioural traits and of their use for welfare assessment.