On October 9, 2006, a seismic event (m b 3.9) occurred in North Korea, which was later announced as a nuclear test. We studied the seismic event from the viewpoint of seismological discrimination without a priori information of the source. This paper presents a compound discriminant method that enhances the ability to distinguish between earthquakes and explosions on the Korean Peninsula. The method includes four single discriminants based on seismic spectral amplitude ratios. These discriminants are used as single variables in a multivariate statistical analysis to derive a compound linear discriminant function that optimally separates the two populations. Short-period vertical component seismograms recorded by seismo-acoustic array (CHNAR) are used for the method. Combining the mutually complementary discrimination abilities of the single methods improves the discriminant power and lowers the probability of misclassification to 1.7% for the data sets. Using this compound method, we were able to identify the North Korea seismic event as explosion-induced. This discrimination technique can be applied automatically to new observations and extended to other seismic arrays or networks. The combination of these seismic discriminants with infrasound observations will increase our ability to detect surface explosions on the Korean Peninsula.