Abstract We re-examine the utility of teleseismic seismic complexity discriminants in a multivariate setting using United Kingdom array data. We measure a complexity discriminant taken on array beams by simply taking the logarithm of the ratio of the P-wave coda signal to that of the first arriving direct P wave (βCF). The single station complexity discriminant shows marginal performance with shallow earthquakes having more complex signatures than those from explosions or deep earthquakes. Inclusion of secondary phases in the coda window can also degrade performance. However, performance improves markedly when two-station complexity discriminants are formed showing false alarm rates similar to those observed for network mb − Ms. This suggests that multistation complexity discriminants may ameliorate some of the problems associated with mb – Ms discrimination at lower magnitudes. Additionally, when complexity discriminants are combined with mb – Ms there is a tendency for explosions, shallow earthquakes and deep earthquakes to form three distinct populations. Thus, complexity discriminants may follow a logic that is similar to mb – Ms in terms of the separation of shallow earthquakes from nuclear explosions, although the underlying physics of the two discriminants is significantly different.