A method of identification of nonlinear dynamic systems, based on the models constructed 
as Volterra series with the use of pulse test signals, is proposed. 
To separate the response into partial components, the method based on differentiation 
of the target output signal with respect to the parameter - amplitude is applied. 
The errors of the identification method are analyzed. By the example of modeling 
a system with nonlinear feedback, the dependences of identification errors 
on the test signal amplitude are obtained. To increase the noise stability 
of the identification method, the smoothing of estimations of the Volterra kernels 
with the help of the wavelet transforms is used.