Abstract: In GPS/INS integrated navigation,the number of observations is usually less than that of the state parameters,and the single adaptive factor is usually applied in Kalman filtering,which can lead to precision loss of indirect observational parameters.A new algorithm of classified adaptive filtering is presented based on predicted residuals and selecting weight filtering,and the corresponding formulas are given.Finally,an actual calculation is given.The new algorithm can not only degrade the influence of the disturbances from the state but also avoid the loss of estimated precision of indirect observational parameters,and improve the accuracy of the navigation further.