Global conservation strategies commonly assume that different taxonomic groups show congruent geographical patterns of diversity, and that the distribution of extinction-prone species in one group can therefore act as a surrogate for vulnerable species in other groups when conservation decisions are being made. The validity of these assumptions remains unclear, however, because previous tests have been limited in both geographical and taxonomic extent. Here we use a database on the global distribution of 19,349 living bird, mammal and amphibian species to show that, although the distribution of overall species richness is very similar among these groups, congruence in the distribution of rare and threatened species is markedly lower. Congruence is especially low among the very rarest species. Cross-taxon congruence is also highly scale dependent, being particularly low at the finer spatial resolutions relevant to real protected areas. 'Hotspots' of rarity and threat are therefore largely non-overlapping across groups, as are areas chosen to maximize species complementarity. Overall, our results indicate that 'silver-bullet' conservation strategies alone will not deliver efficient conservation solutions. Instead, priority areas for biodiversity conservation must be based on high-resolution data from multiple taxa.