Virtual fences (VF) are based on sensory cues delivered to domestic animals with the aim of controlling their spatial behaviour without any physical barrier. VF represent an approach for controlling the distribution of free-ranging large herbivores in large rangeland areas where the implementation of human control or the use of physical fencing is made impossible by their cost, environmental concerns or specific regulations. The potential of using VF for sheep was investigated by studying the spatial behaviour of groups of 5 or 32 ewes submitted to a commercial animal-borne VF device aimed at containing them in a defined area in a series of experiments. All tests were performed in pasture alleys virtually divided in their centre by a 4-m-wide warning/punishment zone). The ewes were introduced on one side of the zone, the other side being made attractive by the presence of tall grass (feed attractant), the presence of peers (social attractant) or being made unattractive by the presence of short grass (same feed as in the alley). In a first experiment, ewes were trained in groups of 5 with a feed attractant, following a 3-stage procedure: 30-min adaptation, 90-min training with a visual cue (three white strings above the wire of the VF) and 30-min test without a visual cue. In a second experiment, the test stage was repeated for each group with feed, social and zero attractants. In a third experiment, the test with feed attractant was performed with groups of 32 ewes comprising previously trained and equipped ewes mixed in different percentages of trained: naïve (untrained and unequipped) ewes (100 : 0, 75 : 25, 50 : 50 and 0 : 100%). One year later, this test was repeated with a feed attractant either close (30 m) or far (60 m) from the punishment zone. By increasing the distance to the feed attractant, it was anticipated that this would stimulate the conflict of motivation between feed and peers. Ewes learned readily the concept of VF after a couple of contacts with