A suitable virtual fencing system could improve pasture management and decrease labour and costs. However, recent research has shown that cows need training in order to be able to understand the virtual fencing mechanism. To improve the understanding of the cows it was envisaged to develop a new approach that supports the learning process. As cows rely more on visual cues than on audio signals, we set out to look into the possibility to use visual cues at least for training purposes. A first prototype with dual warning signals, based on audio and visual cues was developed. The idea behind it was to form a cue consequence association between the audio cue and the electric stimulus by temporarily supporting the learning efforts using visual cues. Therefore, a system based on a raspberry pi platform was developed. As a network and communication with a higher number of devices is paramount, the message queuing telemetry transport protocol was used. This protocol is particularly suitable for machine-to-machine communication through the ‘publish/subscribe’ functionality. Based on theoretical considerations and pretests, a possible luminaire design was created and simulated with a simulation software. A first prototype was built and tested under different light conditions. The system has shown initial promise. However, as the visual cues do not work in sunlight, the next step will be to develop an automated learning system, which could be used between dusk and dawn.