An automated method for removing line spectrum elements embedded in colored spectra is presented. Since smooth spectrum estimates are desired, line spectra tend to smear out over an effective smoothing window. This induces a bias in spectrum estimation that seriously degrades signal-to-noise ratios, spectral deconvolution or any other operation where spectrum shape is important. Multitaper analysis provides a simple algorithmic solution including a method of determining where spectral peaks are both significant with high power. The method is completely general, and examples include estimation of signal-to-noise ratio at the 1990 high frequency array, Pinyon Flat, CA. A comparison of noise spectra line segments and signal spectra line spectra reveals similarities associated with instrument noise and shallow resonances stimulated by incoming seismic signals. Identification and removal of resonances can provide a better means of estimating background noise spectra for modeling earthquake source spectra and path effects associated with attenuation. ¿American Geophysical Union 1995