Social relationships are important in social species. These relationships, based on repeated interactions, define each partner’s expectations during the following encounters. The creation of a relationship implies high social cognitive abilities which require that each partner is able to associate the positive or negative content of an interaction with a specific partner and to recall this association. In this study, we tested the effects of repeated interactions in a training context on the memory kept by 23 young horses about humans, after 6 and 8 months of separation. They were divided in two groups: horses trained to remain immobile on a vocal command with positive reinforcement, which was a food reward (PR, N=11) and controls, trained by simple repetition of the task with no reinforcement, nor punishment (C, N=12). After learning the immobility command, horses from both groups underwent several handling procedures using the same command. Horses trained with the reward required less time to complete training than controls (XPR=3.75±0.08 h, XC=5.24±0.24, MW, P<0.001) and six months after the end of training, they remained immobile for longer than controls when given the vocal order (XPR=55.8±2.2 s, XC=38.0±6.3, MW, P<0.05). The association of a reward with a learning task in the training context induced positive reactions towards humans (e.g. sniffing the trainer: XPR=95.5 ±14.4 occurrences, XC=57.5 ±.8.5, MW, P<0.05). It also increased contact and interest, not only just after training, but also several months later (Time spent close to the trainer 6 months after training: XPR=59.8±4.6%, XC=15±5.6, MW, P<0.001), despite no further interaction with humans. In addition, this ‘positive memory’ of humans extended to novel persons (Time spent close to an unknown person: XPR=39±6.4%, XC=7.2±3.2, MW, P<0.01). Overall, these findings suggest remarkable social cognitive abilities that can be transposed from intraspecific to interspecific social contexts.