Abstract Short-period regional phases play an important role in identifying low-magnitude seismic events in the context of monitoring the Comprehensive Test Ban Treaty. Amplitude ratios of regional phases comprised mainly of P-wave energy (Pn, Pg) to those comprised mainly of S-wave energy (Sn, Lg) effectively discriminate between explosions and earthquakes in many regions, particularly at frequencies higher than 3 Hz. At lower frequencies, discrimination is usually poor due to large scatter that causes overlapping of event populations. Scatter in regional discriminant measures such as Pg/Lg ratios is caused by both source and propagation effects, and reducing the scatter imparted by the latter is essential to improving the discriminant performance when events do not share identical paths. Regional phases experience distance-dependent amplitude variations due to effects such as critical angle amplification, geometric spreading, and attenuation. Discriminant measures are usually corrected for empirically determined distance trends for a given region, but large scatter persists after such corrections. This study seeks to develop more sophisticated empirical corrections for path properties in order to further reduce the scatter in regional discriminant measures caused by propagation effects. Broadband seismic waveforms recorded at station WMQ, in western China, demonstrate that regional Pg/Lg ratios show significant distance dependence for frequencies less than 6 Hz. However, variations in crustal structure cause additional path-specific amplitude fluctuations that are not accounted for by regionally averaged distance corrections. Blockage of Lg phases on paths traversing the margins of the Tibetan Plateau is one such effect. Regression analysis demonstrates that Pg/Lg ratios measured at WMQ display significant correlations with path-specific properties such as mean elevation, topographic roughness, basement depth, and crustal thickness. Multiple regressions using optimal combinations of parameters yield corrections that reduce variance in Pg/Lg measurements for frequencies less than 3 Hz by a factor of 2 or more relative to standard distance corrections. This should systematically enhance the performance of the Pg/Lg discriminant at low frequencies. The method presented here can be used for all regions and all short-period regional discriminants. It is likely that the extraordinary crustal heterogeneity in western China represents an extreme case of path-dependent effects.