In this paper, we present a modelling of an eye motion tracking system. The system consists of an moving magnet and three static magnetic sensors, which implies the magnet embedded in a contact lens and sensors fixed on the spectacles in the application scenario. When the eye is moving, the changing relative position between sensors and magnet will result in different sensory outputs that encodes eye movement information. The simulation of eye movements and corresponding magnetic fields was carried out in MATLAB MathWorks software. After collecting the sensory output, artificial neural network was used to decode the signal and classify the direction of gaze. The gaze was classified with 9 or 17 regions, three levels of noise and 5 positions of the sensors. Each test was run on the same conditions and was repeated 10 times for repeatability. It was found that 3 sensors together achieved the best accuracy for noisy signals with 96.1% and 93.7% for mid and high noise, respectively. The 17-label variant showed below 90% for both mid and high level noise, proving to be not suitable for the application. Based on this simulation, better tracking method can be derived.