Modern day structural health monitoring involves prediction of structural health for possible future load cases for which structure may behave nonlinearly and thus rendering its simplistic linear predictor model obsolete. Among the existing nonlinear material models Bouc-Wen hysteresis model drew most of the attention in recent past due to its wide applicability for different material hysteresis types and ease of implementation. The accuracy of the response predicted by this model entirely depends on how correctly the model parameters have been selected. However due to the inherent nonlinearity and complexity in the model existing parameter identification algorithms are not always certain to produce exact parameter values using limited computational resources. In this article we demonstrate a new technique based on Extended Kalman filtering with adaptive selection of state and measurement covariance matrix to identify parameters of the nonlinear material model with the objective to reduce computational expense. Identification is performed using two different methods: first, in the “iterative” approach, in each iteration step the nonlinear model with estimated parameter is simulated and response is considered to be erroneous measurement which needs to be filtered, whereas, in the second (“sequential”) approach, at each time step current estimate of parameter is used to simulate the model to predict response for next time step and subsequently filtering is performed in real time and parameters are updated at each time step i.e. online. Pros and cons for both these methods are discussed and a conclusive suggestion has been given based on its field of applicability.