Abstract: In agriculture machinery navigation system, Kalman filter is widely used, but it is not good for non-linear system. An improved particles filter methodology used in data fusion was discussed to fuse the data for the better position information. The method involves three parts, the first is anti-outlie step to reduce the error effectively made by GPS; the second is to induct the UKF as the important function, and resample to avoid the degradation; the third is to use MCMC step to reduce the sample impoverishment. The simulation results show that the improved particle filter method can increase the accuracy and reduce the position error. The method can meet the requirements of agriculture machinery navigation.