Abstract: The aim of this study was to test the existence of a “reactivity-to-humans” trait which is stable over time and across situations. For this purpose, four test situations were repeated on the same animals at 8 months, 1.5 years (N = 110 horses) and 2.5 years of age (N = 54 horses). These situations involved a “familiar passive human test” during which a familiar person stayed motionless in the test pen, a “familiar active human test” and an “unfamiliar active human test” during which a familiar or unfamiliar person attempted to touch the horse, and a “halter-fitting and heart-rate measurement test”, during which the horse was haltered and equipped with a surcingle and then had its heart-rate measured. Many behavioural variables were recorded during the tests, and correlations between ages and variables measured during other situations were investigated. When correlations were found, we concluded that the variables corresponded to temperament traits which are stable over time and across situations. The horses tested were divided into four groups according to breed (Anglo-Arab and Welsh) and year of birth (2001 and 2002). Data for each group were analysed separately with Spearman rank correlations. Regarding stability over time, there was a strong correlation between ages in the frequency of licking/nibbling the passive human, the time taken by a familiar or unfamiliar person to touch the horse, and the time taken to fit a halter (e.g. the time taken to touch the horse's muzzle was significantly correlated between ages in 7 out of the 8 possible cases: 0.40 <= R <= 0.67). Regarding stability across situations, results indicate that the more frequently a horse sniffed, licked or nibbled a passive human, the easier it was for either a familiar or unfamiliar active human to touch and halter it. Moreover, a horse which could be touched easily by a familiar human could also be touched and haltered easily by an unfamiliar human. These behaviours, which are linked directly to humans, therefore indicate good stability over time and across situations. In conclusion, this stability suggests the existence of a “reactivity-to-humans” temperament trait, whether the human is passive or active, familiar or unfamiliar. When the animals do not have frequent contact with humans, this reactivity is stable over time and can be measured as early as 8 months of age using the behavioural variables cited above.