A linear model of signal and noise processes generated during electron spectrum measurements is proposed. The relationships are derived for estimation of signal and noise spectral power densities which are used in the expression for the stationary optimal Wiener filter applied for spectrum deconvolution to separate overlapping peaks. A nonstationary version of the optimal Wiener filter is suggested and formulas for its parameters calculation are obtained. Numerical modeling results have shown that the non-stationary filter leads to significant resolution enhancement and increase in reliability of results due to substantial residual oscillation of the solution, that may result in spurious spectral peak identification.