The influence of the variance of the presupposed measured error is analyzed briefly. A new method is presented on the basis of the on-line estimation for the variance of the measured error through spatial-temporal synthetical analysis. It can make full use of redundancy and complementary in- formation, which comes from the single sampled data of multi-sensors and the multi-times sampled data of single sensor, thereby, which improves the influence of modeling error caused by the shortage of prior information for the condition of observed. At the same time, utilizing the characteristic that filtering error covariance expresses filtering precision, the dynamic and reasonable distribution of the every sensor weight coefficient is accomplished. Finally,a new algorithm of multi-sensors fusion based on the variance of the measured error adaptive is given.Monte-Carlo simulation shows that this algorithm can obviously improve the efficiency of maneuvering target tracking.