Mass extinction events are recognized by increases in extinction rate and magnitude and, often, by changes in the selectivity of extinction. When considering the selective fingerprint of a particular event, not all taxon extinctions are equally informative: some would be expected even under a ‘background’ selectivity regime, whereas others would not and thus require special explanation. When evaluating possible drivers for the extinction event, the latter group is of particular interest. Here, we introduce a simple method for identifying these most surprising victims of extinction events by training models on background extinction intervals and using these models to make per-taxon assessments of ‘expected’ risk during the extinction interval. As an example, we examine brachiopod genus extinctions during the Late Ordovician Mass Extinction and show that extinction of genera in the deep-water ‘Foliomena fauna’ was particularly unexpected given preceding Late Ordovician extinction patterns.