Recent studies have demonstrated that horses can recognize humans based simply on visual information. However, none of these studies have investigated whether this involves the recognition of the face itself, or simply identifying people from non-complex external clues, such as hair color. To go beyond this we wanted to know whether certain features of the face were indispensable for this recognition (e.g., colors, hair or eyes). The 11 horses in this study had previously learned to identify four unfamiliar faces (portrait view and in color) presented repeatedly on a screen. We thus assessed whether they were able to identify these same faces spontaneously when they were presented in four other conditions: profile view, black and white, eyes hidden, changed hairstyle. The horses' performances remained higher than chance level for all the conditions. In a choice test under real conditions, they then approached the people whose face they had learned more often than unknown people. In conclusion, when considering all the individuals studied, no single facial element that we tested appears to be essential for recognition, suggesting holistic processing in face recognition. That means horses do not base their recognition solely on an easy clue such as hair color. They can also link faces from photographs with people in real life, indicating that horses do not process images of faces as simple abstract shapes.