Exactness of the Bouc-Wen hysteresis model entirely depends on the correctness of the model parameters. This paper applies Extended and Unscented Kalman filtering approach with adaptive process and measurement error covariance matrix to identify these parameters in an efficient way. We define time invariant model parameters as states of the process while the error in model output is defined through the measurement equation. In addition we compare two different methods of identification of the hysteresis parameters based on their computational cost and convergence criteria and their fields of applicability are discussed. First, in the “iterative” approach, parameter updating is done iteratively comparing model output to measured response for a fixed time span. While in the second (“sequential”) approach, updating is performed in real time comparing true and model response in each time step. Two numerical cases are investigated: a SDOF system and a three story shear frame building for validation.