We present an adaptive automated algorithm for estimating the spectral decay of high frequency waves from recorded acceleration time histories of strong ground motion, along with a simple P- and S-phase picking algorithm to achieve complete automation. Our approach negotiates site resonance peaks and spectral corners through a spectral linearity criterion that on average provides indistinguishable results compared to manual estimates. The overall spectral decay, represented by the so-called “kappa” (κ) parameter, is demonstrated on a dataset of accelerograms from earthquakes of magnitudes 3.7-6.5 at distances of 1–76 km recorded on site conditions classified as rock and stiff soil, respectively, in South Iceland. The automatic procedure gives an average estimate of κ¯=37.2±13.6 ms on rock. The data did not allow the robust determination of linear distance dependence, nor a distinction of κ between the two site classes due to the generally large scatter of κ values. Using a subset of stations for the analysis however, a slight distance-dependence could be observed but is likely due to the influence of the quality factor being proportional to frequency. The results indicate that source and site effects drive κ values in South Iceland, and that a formal inclusion of source contributions to the parametrization and analysis of the spectral decay is needed.