Aimed at the problems that the unknown statistical feature of measurement noise affects the filtering precision of integrated navigation, an interacting multiple model algorithm based on improved innovation adaptive filter is proposed. In the process of estimating the innovation covariance, a set of estimation windows with different lengths is used to optimize the selection of the estimation window. Then the measurement noise covariance is estimated, taking this estimated value as a centre, the models of interacting multiple model algorithm are built symmetrically, and the subfilters in the algorithm is fed. The contradiction between the number of model and the computing speed is solved by using this algorithm. The simulation results show that this method is high in accuracy and strong in antiinterference ability, compared with the standard Kalman filter algorithm and the traditional innovationbased adaptive multiple model method.