In many engineering problems, the power density spectra of random signals, with highly periodic components, have to be measured, very often this problem is reduced to the one of identifying the unknown power spectrum with a rational function. The identification techniques differ from each other only by the statistical criteria of the procedures used to estimate the assumed model parameters. In this work three spectral analysis methods are examined: Maximum Entropy, Minumum Residual, and that proposed by Hsia and Landgrebe. The power spectrum model assumed by the first method is a rational function , for the other two methods, has poles and zeroes. The Minimum Residual method uses, when the process under study is Gaussian, the maximum likelihood estimation principle. The method developed by Hsia and Landgrebe has a estimation procedure simpler, but is statistically less significant. This work investigates the possibility of substituing the Maximum Entropy model by another one, with poles and zeroes, but with less parameters to be estimated and better performace in the estimation of power spectra that contain peaks. This investigation is made by computer simulation and the efficiency of the methods is compared in terms of localization and resolution of the peaks contained in the power spectra.