The goal of performing spectral decomposition is to determine the frequency spectra of the seismic signal as a function of traveltime. In conventional short-window Fourier methods and in continuous-wavelet transforms, there is a trade-off between temporal resolution and frequency resolution. By using a new method, spectral decomposition can be obtained at the time resolution of the input trace, which is independent of frequency. Furthermore, this decomposition has a constant frequency resolution on the order of a few hertz. However, there still might be some trade-offs. To perform this decomposition, a sliding calculation window is required, the length of which has some effect on the results. Long windows produce higher-frequency resolution, as is typical in performing Fourier transforms. However, higher resolution does not necessarily mean more accurate results. Different window lengths can produce different-looking (sometimes significantly different) spectral tuning curves because of the interrelationship of the reflections contained in the window. Guidelines for the sensitivity to the calculation window length have not yet been determined. This new spectral-decomposition approach is illustrated using four simple synthetic models and a line from a widely used data set.