The paper presents a method of modal parameter estimation based on RLS (Recursive Least Square) algorithm, and wavelet filtering. The wavelet filtering gives possibility to decoupling frequency components of signal response of structure. This operation can also reduce the order of the signal model estimated by RLS algorithm. An additional advantage of this method is the possibility of adapting the wavelet filter parameters to the changing parameters of the system. Reduced model order significantly reduces the time of estimation of modal parameters, which enables the real – time implementation of the method. Due to recursively updated covariance matrix of model parameters, the confidence intervals of modal parameters can be also estimated. All routines have been implemented and tested in MATLAB®. The method have been tested on simulated data delivered by an AIRBUS team and on the test bed with a variable stiffness.