The paper presents results of comparative study of different methods for estimation of continuous energy spectra of random returns in coherent pulse radars. We analyze these methods’ “extreme” capabilities on reconstruction of continuous spectra under hypothetic conditions of known statistical characteristics of input signals. We investigate also spectral estimation methods’ “real” capabilities under real-world operational conditions with parametric a priori uncertainty and small-size samples being used to estimate a priori unknown parameters of reflections. The causes of known drawbacks of classical (nonparametric) spectral estimation (SE) methods are discussed. Then we consider parametric SE methods, sources of their drawbacks and ways to overcome them.