In conventional feedback filter, the feedback from some unobservable system states will lead to the divergence of navigation errors. Further the measurement equation of global positioning system/strapdown inertial navigation system (GPS/SINS) is nonlinear, and it increases the complexity of the filter. To solve these problems, a linear measurement equation was derived in this paper. As a complement to this measurement model, a composed adaptive revising filtering method based on the observability of states was presented as well. The eigenvalues and eigenvectors analysis method of error covariance matrix was chosen to analyze the observability of system states as a result of the comparison of three popular observability analysis methods. The feedback factors were thus designed according to the analyzed results. As a consequence, errors from the original data of SINS and GPS receiver can be corrected. Simulation results demonstrate that our proposed method is effective and can improve the estimate accuracy of the system in incomplete observable condition.