Mass extinctions are important to macroevolution not only because they involve a sharp increase in extinction intensity over “background” levels, but also because they bring a change in extinction selectivity, and these quantitative and qualitative shifts set the stage for evolutionary recoveries. The set of extinction intensities for all stratigraphic stages appears to fall into a single right-skewed distribution, but this apparent continuity may derive from failure to factor out the well-known secular trend in background extinction: high early Paleozoic rates fill in the gap between later background extinction and the major mass extinctions. In any case, the failure of many organism-, species-, and clade-level traits to predict survivorship during mass extinctions is a more important challenge to the extrapolationist premise that all macroevolutionary processes are simply smooth extensions of microevolution. Although a variety of factors have been found to correlate with taxon survivorship for particular extinction events, the most pervasive effect involves geographic range at the clade level, an emergent property independent of the range sizes of constituent species. Such differential extinction would impose “nonconstructive selectivity,” in which survivorship is unrelated to many organismic traits but is not strictly random. It also implies that correlations among taxon attributes may obscure causation, and even the focal level of selection, in the survival of a trait or clade, for example when widespread taxa within a major group tend to have particular body sizes, trophic habits, or metabolic rates. Survivorship patterns will also be sensitive to the inexact correlations of taxonomic, morphological, and functional diversity, to phylogenetically nonrandom extinction, and to the topology of evolutionary trees. Evolutionary recoveries may be as important as the extinction events themselves in shaping the long-term trajectories of individual clades and permitting once-marginal groups to diversify, but we know little about sorting processes during recovery intervals. However, both empirical extrapolationism (where outcomes can be predicted from observation of pre- or post-extinction patterns) and theoretical extrapolationism (where mechanisms reside exclusively at the level of organisms within populations) evidently fail during mass extinctions and their evolutionary aftermath. This does not mean that conventional natural selection was inoperative during mass extinctions, but that many features that promoted survivorship during background times were superseded as predictive factors by higher-level attributes. Many intriguing issues remain, including the generality of survivorship rules across extinction events; the potential for gradational changes in selectivity patterns with extinction intensity or the volatility of target clades; the heritability of clade-level traits; the macroevolutionary consequences of the inexact correlations between taxonomic, morphological, and functional diversity; the factors governing the dynamics and outcome of recoveries; and the spatial fabric of extinctions and recoveries. The detection of general survivorship rules—including the disappearance of many patterns evident during background times—demonstrates that studies of mass extinctions and recovery can contribute substantially to evolutionary theory.