Temperament is a trait of interest as it impacts ease of handling, production and reactivity to stressful situations. Excessive behavioural reactivity, as an aspect of temperament, may lead to reduced welfare and stockperson safety. Current tests of behavioural reactivity involve isolating the subject and it can be difficult to differentiate between behavioural responses. We therefore aimed to develop a startle test in sheep which quantified variability in behavioural reactivity whilst addressing the effects of social isolation. The startle response is a consistent universal reaction to sudden and intense stimulation, however, magnitude of startle can be highly variable and may be modulated by affective state. In this study, 60 ewes were allocated into three social groups: isolated; surrounded by six life-sized pictures of sheep or; three conspecifics penned on two sides (total six). Sheep were tested over two days in sets of 10 per treatment. Sheep were confined to a 2.1 m × 1.2 m area, with a bowl of feed at one end, for four minutes (min). Approximately 15 s after the sheep commenced eating, a sudden one sec blast of compressed air was delivered to the face. Tri-axial accelerometers, visual scoring, agitometer scores and body temperature were used to assess the magnitude of the startle response and the pre- and post-startle fear behaviours. In the 15 s pre-startle, isolated sheep spent less time eating (p < 0.001; mean percentage of time: 37.4% vs 71.2% and 83.3%) and were 2-4 times longer in a vigilant state (p < 0.001; mean percentage of time: 38.9% vs 18.2% and 11.3%) than those with pictures and conspecifics respectively. Sheep in the conspecific treatment were 3-4 times faster to return to eating after startle than those in the isolated and picture treatments and there was a tendency for them to spend more time eating (p = 0.09) and less time vigilant (p = 0.09) post-startle. For each measure of startle magnitude (startle force, startle duration or retreat distance) the effect of treatment was not statistically important (p > 0.22), however each measure was positively correlated to time spent vigilant post-startle and there were significant increases in time to return to eating for every one unit increase in each of the startle magnitude measures independent of social treatment. The use of conspecifics did not provide social facilitation or buffering of the startle response. This shows promise as a behavioural reactivity test which can utilise conspecifics to reduce any potential negative effects of social isolation during testing.