Evolutionary bet hedging encapsulates the counterintuitive idea that organisms evolve traits that reduce short-term reproductive success in favor of longer-term risk reduction. It has been widely investigated theoretically, and many putative examples have been cited including practical ones such as the dormancy involved in microbe and weed persistence. However, long-term data on demographic variation from the actual evolutionarily relevant environments have been unavailable to test for its mechanistic relationship to alleged bet hedging traits. I report an association between delayed germination (a bet hedging trait) and risk using a 22-year data set on demographic variation for 10 species of desert annual plants. Species with greater variation in reproductive success (per capita survival from germination to reproduction x per capita fecundity of survivors) were found to have lower average germination fractions. This provides a definitive test using realistic data on demographic variance that confirms the life history prediction for bet hedging. I also showed that the species with greater long-term demographic variation tended to be the ones with greater sensitivity of reproductive success to variation among years in growing-season precipitation.