By combining the chisquare test and gray prediction, an algorithm of hard fault detection and isolation was introduced for the navigation of agricultural robot. The Kalman filtered data and historical valid GPS data were applied to detect and isolate positioning fault caused by the GPS failure, and none of other sensors were applied. With the prototype of the agricultural wheeled mobile robot, the experimental results showed that the GPS positioning fault was isolated effectively and had no any effects on the robot navigation. The error of gray prediction of GPS positioning data were less than 10cm, and the missing rate of fault detection was reduced further when the false alarm rate of the chisquare test method was only 0.05.