Abstract: GPS is applied widely in autonomous navigation of the agricultural machinery. Its positioning error, however, is characterized by autocorrelation, can not satisfy the requirement of Kalman filtering, which is the base of the integrated navigation system of the agricultural machinery. So the characteristic of GPS positioning error was described as AR model with the time series analysis. Then the method to predict and modify the GPS positioning error with AR model and optimal estimation of Kalman filtering was introduced. And the corrected GPS positioning data were applied in the Kalman filtering for the integrated navigation of the agricultural machinery. The experimental results showed that the autocorrelation between neighboring positioning error data was decreased dramatically, and being similar to the white noise, no matter the GPS receiver was static or not. And when the tracked path was straight and curve, the maximum tracking error was about 0.15 m and 0.3 m respectively. This method can provide a viable way to achieve high-accuracy navigation with low-accuracy GPS for the agricultural machinery.