I use seismic data from portable digital stations and the broadband Terrascope network in southern California to investigate radiated earthquake source spectra and discuss the results in light of previous studies on both static stress drop and apparent stress. Applying the empirical Green's function (EGF) method to two sets of M 4–6.1 events, I obtain deconvolved source-spectra estimates and corner frequencies. The results are consistent with an ω2 source model and constant Brune stress drop. However, consideration of the raw spectral shapes of the largest events provides evidence for a high-frequency decay more shallow than ω2. The intermediate (≈f–1) slope cannot be explained plausibly with attenuation or site effects and is qualitatively consistent with a model incorporating directivity effects and a fractional stress-drop rupture process, as suggested by Haddon (1996). However, the results obtained in this study are not consistent with the model of Haddon (1996) in that the intermediate slope is not revealed with EGF analysis. This could reflect either bandwidth limitations inherent in EGF analysis or perhaps a rupture process that is not self-similar. I show that a model with an intermediate spectral decay can also reconcile the apparent discrepancy between the scaling of static stress drop and that of apparent stress drop for moderate-to-large events.