Delay-fired mine blasts, which consist of a series of individual shots arranged in a grid pattern and detonated in sequence, can introduce spectral modulations into recorded seismograms. We can exploit spectral modulations to separate delay-fired mine blasts from the remaining event population, which includes single- fired mine blasts and earthquakes. Here, we enhance an existing algorithm (Hedlin, 1998) for the automatic discrimination of delay-fired mine blasts. A total of seven separate discriminants are computed, based on the spectrograms of recorded events. A feature-selection procedure is used to ensure that each discriminant is significant and contributes to the overall performance of the discrimination algorithm. The effect of input parameters on the methodology is explored. The choice of input parameters is made to maximize the mean Mahalanobis distance between the earthquake and delay-fired mine-blast populations. The technique is then applied to a dataset consisting of regional earthquakes and delay-fired mine blasts recorded at a station in Wyoming. The results show that the larger delay-fired mine blasts, the cast blasts, can be identified successfully by using this technique. The smaller mine blasts are not identified with this technique, although such events are of less interest in a nuclear-monitoring perspective. In a drop-one test, 89.5% of the events studied are successfully identified. Of the events that are misclassified, one is a cast blast and seven are earthquakes. The cast blast is misclassified because of noise on one component, which biased the value of a single discriminant. The earthquakes are misclassified because of a greater variance of the seven discriminants for the mine-blast population. The results suggest that this methodology is very successful at identifying cast blasts in Wyoming, and would be an extremely useful method to use as part of an integrated set of discriminants for the identification of small-magnitude regional events.