[en] System identification allows to build mathematical models of a dynamic system based on measured data. System identification is carried out by adjusting parameters within a given model until its output coincides as well as possible with the measured output. The aim of this study is to investigate and model the behavior of complex vibratory systems on the basis of measured excitation and response. The first part of the study describes the theory used in the analysis and the software tools used in the analysis. The second part of the study describes the investigation and modeling of the response of single degree of freedom oscillator excited by sinusoidal and blast excitation. In the third part of the study the system identification of the Kozloduy NPP unit 5 reactor building and Paks NPP unit 1 reactor building is studied and the models are estimated using the method of segmentation of excitation and response. System identification is carried out using MATLAB software by adjusting parameters within a given model until its output coincides as well as possible with the measured output. The types of models used for the were: l) ARX models; 2) ARMAX model; 3) Output-Error (OE) models; 4) Box-Jenkins (BJ) models; 5) State-space models. The model coefficients for different models were calculated using the least-squares and maximum likelihood estimation methods available in MATLAB system identification toolbox. Excitation was in both Paks and Kozloduy case the measured free-field excitation and responses were the vibration responses of the building on the foundation slab level and top of the building. By examining the established models the frequency characteristics of vibration systems were determined with 95 % accuracy and the amplitude response with 80 % accuracy. In case of the steady state response of sinusoidally excited single dof oscillator the modelling gave almost exact results. But in the case of the blast response of the reactor building the obtaining of the accuracy reported in last paragraph required segmentation of the 5 second blast response to 0.5 s segments better results results could be obtained if the segmentation was carried out to the level of 0.1 s segments but this required excessive amount of work and even in this case the agreement was not perfect. In conclusion the applicability of the available software tools of system identification to complex vibratory systems give on satisfactory results even with great numerical effort applied. (orig.)