Direct estimates of extinction risk are frequently unavailable, so surrogate indicators are used in threatened species categorizations, but there is inadequate information on which best predict vulnerability. We compared the ability of 16 frequently-used factors to predict extinction risk for 45 vertebrate taxa. Median times to extinction were estimated using taxon-specific stochastic population models. Population size (N) and trend were clearly the best correlates of extinction risk in our data set. Stepwise multiple regression with additive and interaction terms identified N, N × trend, plus taxonomic level, number of sub-populations × N × trend, number of offspring (O) and N × O as predictors, and explained 70% of the variation. Trend was important in large, but not in small populations. Population size is the most important data to collect for threatened species and with trend should be the major focus in endangered species categorization and state of the environment reporting.