Learning through the observation of others allows the transfer of information without the costs incurred during individual trial and error learning. Horses (Equus caballus) are a highly social species, which might be expected to be capable of learning from others, but experimental findings are inconsistent, and potentially confounded by social facilitation effects not related directly to the learning of the task. We refined the methods used in previous equine social learning studies, to examine and distinguish specific social influences on learning of a task: we used predefined group leaders rather than agonistically dominant individuals to demonstrate a detour task to familiar conspecific observers; in addition we had two control groups: a non-observer (true control) and a group with the demonstrator simply present at the goal (social facilitation control). 44 socially kept horses were allocated to one of the three test conditions and took part in five trials each. Success rate, latency and detour direction were recorded. There was no significant difference between the three groups in the likelihood of them succeeding in the task nor latency to succeed; however there was a significant difference in the route chosen by the groups, with the true control choosing the side with the entrance gate significantly more than either the observer group or social facilitation group. Both of the latter two groups chose to go in the same direction relative to themselves, regardless of which side the gate was. Seven out of nine horses in the observer group chose the same direction as their demonstrator every time. Our results show a significant role of social facilitation on detour behaviour and highlight the importance of including adequate controls for simpler cognitive influences on behaviour before claims can be made about the specific learning of motor actions or goal directed behaviour. Social cues may be important to horses if the task is sufficiently challenging and motivationally important, so future work should consider more demanding, but ecologically relevant situations, in order to maximise the potential revelation of social learning effects which do not depend on simple local or stimulus enhancement effects.