Improving tracking accuracy is one of the most important issues for applying augmented reality to nuclear power plant field work. When employing tracking method using single camera and markers, tracking accuracy depends highly on marker arrangement. Manual design of marker arrangement is difficult to find a marker arrangement with high tracking accuracy. Therefore this study is subject to develop a marker arrangement optimization system based on genetic algorithm. Tracking error computation is necessary for each marker arrangement in marker arrangement optimization. Present tracking error computation method are all probabilistic and therefore feasible to nuclear power plant A wheel tracking error computation algorithm is developed. A marker arrangement optimization algorithm is realized by combining genetic algorithm with the wheel tracking error computation algorithm. Trial results show that tracking accuracy can be improved significantly by applying the advices provided by the marker arrangement optimization system.