A methodology is investigated for assessment of feeding preferences of grazing sheep where satellite tracking is carried out using GPS collar units. Sheep grazing preferences were analysed by a statistically based approach under constraints of spatial error in the GPS tracking data. The statistical significance of grazing locations was determined by testing a null hypothesis on animal locations in two adjacent pastures using the normal approximation to the binomial distribution for proportions. The analysis also included calibration and compensation for the precision of the GPS receiver, which produced an uncertain decision boundary between the paddocks. It was concluded that the approach was effective in dealing with GPS spatial errors and in comparing the feeding preferences of sheep.