We compare P-wave spectra of quarry blasts and earthquakes recorded by the southern California seismic network (SCSN) between 2000 and 2005, with the goal of developing methods to discriminate between these events. We process the spectra using an iterative robust least-squares method to isolate source, receiver, and propagation path contributions. This corrects for first-order attenuation structure, as well as near-receiver site effects and any errors in the instrument response functions. Using the earthquake spectra and a simple ω-2 source model, we compute an empirical Green’s function (EGF) to remove the trade-off between the source terms and other parameters in our model. A constant stress-drop model gives a good fit to the observed average earthquake spectra over a wide range of moment magnitude, but provides a mediocre fit to the average quarry blast spectra, which have a generally steeper fall-off at high frequencies than ω-2. We also compare P- and S-wave amplitudes and find modestly smaller average S amplitudes for the explosions compared to the earthquakes. For southern California, the root-mean-square (rms) misfit of P-wave spectra to an ω-2 source model is a more reliable explosion discriminant than the S-to-P amplitude ratio and works for about 90% of the events.