(1) where 0(m) denotes an additive noise with the mean value M[0(m)] = 0 and the dispersion R. Utilization of estimates of the type (0t ~ ~S (~,,) requires application of statistics and a large duration of the experiment, in the course of which the nonstationary character of the signals investigated manifests itself; this nonstationary character is related to the drift of the constant components of the signals and to the change of the zero levels of the apparatus. Of practical interest for many applications are methods of deriving X(m) from a single realization S(m) in such a way that the values of the divergence are small; the divergence is the important criterion for spectroscopy and indicates the quality of the estimate made, The filters with constant parameters, which are extensively used in the practice of low-frequency smoothing, do not meet the nonstationary character of the signal S(~) observed at the output of the spectrometer and result in distortion of the shape or the parameters of the line examined, Application of special criteria for the synthesis of such filters is possible only in isolated cases and requires a priori knowledge of the shape of the signal itself [I]. In paper [2], a method based on modification of the Kalman filter (KF) is proposed for smoothing nonstationary signals. In the present paper the application of this method in spectroscopy is discussed on the example of separating a signal with the Lorentz shape