Spectral ratio disciminants are applied to 72 Western United States earthquakes and 64 NTS explosions recorded at four broadband seismic stations surrounding NTS. The ratio of the energy in the 1- to 2- and 6- to 8-Hz bands for Pn, Pg, and Lg is calculated and a simple distance correction applied to the data. The spectral ratio appears to have potential as a disciminant at relatively small magnitudes (3.0 < mb < 4.5) and single station misclassification probabilities for detected phases range from 4 to 33 per cent, with Lg showing the best performance followed by Pg and Pn. Below mb 4.5 to 5.0, the earthquakes are observed to have more high-frequency energy than the explosions for all three phases. This observation may be due to actual source differences or to depth-dependent effects of attenuation on the shallow explosions and deeper earthquakes. At higher magnitudes, the two populations merge and discrimination is poor. Overburied explosions are characterized by the existence of more high-frequency energy than those at standard containment depths and are often misclassified. The Brune dislocation source model is shown to satisfactorily fit the earthquake spectral ratios plotted as a function of mb for reasonable stress drops. However, the Mueller-Murphy explosion source model has problems predicting the decrease in the explosion spectral ratio observed at higher magnitudes (mb > 4.5 to 5.0). This may be due to uncertainties in the apparent source-time function for explosions. These complications may be due to changes in the dynamic response of the material in the near-source region as a function of overburden pressure, effects of secondary sources, or to variations in regional phase excitation with depth.