The displacements estimated using seismic accelerations are useful for post-earthquake inspection of structural damage. In previous studies, displacements were usually obtained through non-trivial measurement or double integration of the measured acceleration response and then an additional correction process. However, from the signal processing perspective, double integration induces substantial numerical errors. In this study, a displacement estimation approach is developed based on the extended Kalman filter (EKF) with an embedded Bayesian noise-parameters updating technique. The displacement is estimated by EKF with updated time-varying noise parameters based on a Bouc–Wen model. Numerical simulations and a shaking‐table test are presented to demonstrate the proposed method.