The unmanned operation of agricultural machinery in whole field is an important part of unmanned farms. In order to perform high-efficiency autonomous navigation in field, an improved Stanley path tracking control model based on fuzzy algorithm was proposed, where the control gain was adaptively changed according to the tracking error. Base on the kinematics model and Stanley model, the whole field tracking method of agricultural machinery was studied. Firstly, the membership function was constructed by using lateral deviation and heading deviation as the input variables. Fuzzy inference rules were designed by analyzing and summarizing the experimental data. In order to determine the gain coefficient of the control model, the center of gravity method was used as the defuzzification method, which improved the adaptivity of the Stanley model under complicated path condition. In order to verify the effectiveness of proposed algorithm, a mobile vehicle equipped with GNSS navigation system was used as the experimental platform, where the whole field path tracking of combine harvester was tested and discussed. Experiment results indicated that when the operation speed of straight line was at 2.5m/s, and the turning speed was at 1m/s, the maximum error of whole field path tracking was less than 3cm. When the initial lateral deviation was 3m, the distance of guided trajectory was no more than 5m, which improved the path tracking accuracy of the conventional Stanly model significantly. The proposed algorithm satisfied the need of efficient automatic navigation operations for agricultural machinery in the whole farmland field.