High-frequency regional records from small earthquakes (magnitude <4.5) and comparable magnitude chemical explosions are analyzed to find a reliable seismic discriminant in southern Russia near Kislovodsk. The digital, three-component seismograms recorded during 1992 by the Caucasus Network operated by Lamont-Doherty Earth Observatory since 1991 in the distance ranges 15 to 233 km are used. Mean vertical-component Pg/Lg spectral amplitude ratios in the band 8 to 18 Hz are about 1.3 and 3.2 for earthquakes and explosions, respectively, in this region. We find that the vertical-component Pg/Lg spectral ratio in the frequency band 8 to 18 Hz serves quite well for classifying these events. A linear discriminant function analysis indicates that the Pg/Lg spectral ratio method provides discrimination power with a total misclassification probability of about 7%. The Pg/Lg spectral ratios of rotated, three-component regional records improve the discrimination power of the spectral ratio method over the vertical-component Pg/Lg ratios. Preliminary analysis indicates that distance-corrected vertical-component Pg/Lg ratios improve the discrimination power by about 4% over uncorrected ratios. But we find that an even better discriminant is the Pg/Lg spectral ratio of the three-component regional records corrected for the free-surface effect. In the frequency band 8 to 18 Hz, the free-surface-corrected three-component Pg/Lg spectral ratio provides discrimination power with a total misclassification probability of only 2.6%. Free-surface-corrected and network-averaged Pg/Lg spectral ratios provide transportability of the spectral ratio method to various regions worldwide.