Spectral discrimination methods applied at regional distances are based on the observation that there is a difference in the frequency content of the seismic waves radiated from the source region of earthquakes and explosions. We have investigated a spectral discrimination method that exploits the scattered energy contained in the event codas of seismograms to measure the relative source excitation of earthquakes and explosions as a function of frequency. Key strengths of the coda method include the following: (1) it is not affected by source radiation pattern because the scattering process averages the radiation over the focal sphere, (2) path corrections are not required with the coda method because coda waves are scattered throughout a common crustal volume, (3) it can be successfully applied using data from only a single station, and (4) it can be successfully applied using data with clipped direct arrivals. To implement the coda method, we first estimate a “type curve,” which represents the average shape of all the individual coda decay curves for a particular frequency band. The shape of the type curve (as well as the shape of all individual coda decay curves) is controlled by the scattering and attenuation properties of the crustal volume sampled by the coda waves (a common path effect). The amplitude of the type curve is not defined, and we arbitrarily adjust it to have a zero mean. The amplitude difference between an individual event coda decay curve and the type curve provides a relative measure of the event's source size within the particular frequency band. We define this amplitude difference as the “relative source factor.” Our approach exploits all of an event's usable coda for each spectral measurement, thus providing a more stable measure of relative source size than could be obtained from using a predefined, fixed-length time window. Analysis of 27 Nevada Test Site explosions and 15 southern Great Basin earthquakes (3.1 ≦ M ≦ 4.7), shows that the source factor difference between the 0.5- to 1- and 2- to 4-Hz bands plotted against the 1- to 2-Hz source factor is a stable event discriminant. The method reliably distinguishes small explosions from small earthquakes (3.0 < M < 3.5), even if the explosions are overburied. Because the coda method requires only one seismic station, we believe that it would be especially useful in identifying small, clandestine underground nuclear tests that occur far from established networks.