AbstractWe show that ML–Mc is a viable and regionally portable depth discriminant, and therefore may contribute to nuclear test ban treaty verification. A recent study found that the difference between local magnitude (ML) and coda duration magnitude (Mc) discriminates shallow seismic events (mining blasts, mining‐induced earthquakes, and shallow tectonic earthquakes) from deeper tectonic earthquakes in the Utah region. The shallow seismic events had anomalously high Mc values, with increasingly negative ML–Mc values as depth decreased. Here, we evaluate the performance of ML–Mc as a depth discriminant in three new regions and find that ML–Mc increases between 0 and 9 km depth in all cases. Initially, we investigated ML–Mc as a function of depth for naturally occurring earthquakes in the region around Yellowstone National Park, as recorded by the University of Utah Seismograph Stations. For 3358 Yellowstone earthquakes with well‐constrained depths, we found ML–Mc increased 0.030±0.007 magnitude units (m.u.) for each 1 km increase in depth up to 10 km depth. Next, we examined ML–Mc values for anthropogenic seismicity in northern Oklahoma and southern Kansas, as recorded by the National Earthquake Information Center. For 1628 events with well‐constrained depths, we computed a slope for ML–Mc of 0.022±0.010 m.u./km. Finally, we analyzed ML–Mc for 28,722 well‐located earthquakes in Italy, as recorded by the National Institute of Geophysics and Volcanology, and found an ML–Mc slope of 0.018±0.001 m.u./km. In each case, the quoted error bounds represent 95% confidence regions that exclude zero, implying that the depth dependence of ML–Mc is statistically significant. We performed several robustness tests in which we varied the criterion used to define a well‐constrained depth and the depth range used in the linear fit. In nearly all cases, we found a positive slope for ML–Mc versus depth at a confidence level above 95%.