According to the problem of location failure or large deviation probably generated on account of sensor information loss or abnormal fluctuation when the tractor worked in the complicated field environment, a multi-sensor fusion positioning system based on the weighted-confidence was designed, which composed of global position system and dead reckoning system. The corresponding sub-filter and the global filter were constructed based on the federated Kalman filtering. Then the confidence distance and confidence function of every sensor was defined to compute the confidence coefficient of each sub-filter, which was used as adaptive factor after weighted arithmetic. Finally, the adaptive factors were fused optimally in the global filter to get more accurate position estimates. The computer simulation and field test were carried out to validate the effectiveness and reliability of the designed multi-sensor fusion positioning system.