The purpose of this study was to test the existence of a "fearfulness" trait in horses, by testing the stability across situations and over time of the responses to different fear-eliciting situations. It was also to identify which behavioural parameters are the best indicators of this trait. Sixty-six Welsh ponies and 44 Anglo-Arab horses were successively tested at 8 months and 1.5 years of age. Of these, 33 Welsh ponies and 21 Anglo-Arabs were also tested at 2.5 years of age. At each age, they were subjected to four test situations. The first test involved the introduction of a novel object in the test pen (novel object test). In the second test, a novel area was placed in the pen between the horse and a bucket of food, to determine the time the horse took to cross the area (novel area test). Finally, the third test consisted in suddenly opening an umbrella in front of the horse while it was eating (surprise test). During these tests, many behavioural parameters were recorded. A fourth test consisted of a surprise test during which the horse was held by a handier while its heart rate was measured. Spearman correlations were used to identify links between behavioural parameters measured during different tests and between different ages. Reactions to the first three tests showed consistency between them and over time. Among all the behavioural parameters measured during these tests, sonic presented high stability over time and were well correlated with behaviours expressed during other tests, indicating they are the best indicators of a fearfulness trait: the frequency of licking/nibbling the novel object, the time to put one foot on the novel area and to eat from a bucket placed just behind it, and the flight distance and the time to eat under the opened umbrella. The stability across responses expressed in various fear-eliciting events and over time from 8 months to 2.5 years of age suggests the existence of a 'fearfulness' trait in horses. The different indexes of heart rate measured or calculated during the surprise effect present limited stability over time and almost no correlation with the behavioural parameters measured during the other three tests. We Conclude that, in contrast to the previously mentioned behaviours, these are not reliable measures of it temperament trait. From a practical point of view, this study shows that it is possible to identify a horse's level of fearfulness as early as 8 months of age using the first three tests. (C) 2008 Elsevier B.V. All rights reserved.