The end-Cretaceous mass extinction ranks among the most severe extinctions of all time; however, patterns of extinction and recovery remain incompletely understood. In particular, it is unclear how severe the extinction was, how rapid the recovery was and how sampling biases might affect our understanding of these processes. To better understand terrestrial extinction and recovery and how sampling influences these patterns, we collected data on the occurrence and abundance of fossil mammals to examine mammalian diversity across the K-Pg boundary in North America. Our data show that the extinction was more severe and the recovery more rapid than previously thought. Extinction rates are markedly higher than previously estimated: of 59 species, four survived (93% species extinction, 86% of genera). Survival is correlated with geographic range size and abundance, with widespread, common species tending to survive. This creates a sampling artefact in which rare species are both more vulnerable to extinction and less likely to be recovered, such that the fossil record is inherently biased towards the survivors. The recovery was remarkably rapid. Within 300 000 years, local diversity recovered and regional diversity rose to twice Cretaceous levels, driven by increased endemicity; morphological disparity increased above levels observed in the Cretaceous. The speed of the recovery tends to be obscured by sampling effects; faunas show increased endemicity, such that a rapid, regional increase in diversity and disparity is not seen in geographically restricted studies. Sampling biases that operate against rare taxa appear to obscure the severity of extinction and the pace of recovery across the K-Pg boundary, and similar biases may operate during other extinction events.