Bet hedging is one solution to the problem of an unpredictably variable environment: fitness in the average environment is sacrificed in favor of lower variation in fitness if this leads to higher long-run stochastic mean fitness. While bet hedging is an important concept in evolutionary ecology, empirical evidence that it occurs is scant. Here we evaluate whether bet hedging occurs via seed banking in natural populations of two species of desert evening primroses (Oenothera, Onagraceae), one annual and one perennial. Four years of data on plants and 3 years of data on seeds yielded two transitions for the entire life cycle. One year was exceptionally dry, leading to reproductive failure in the sample areas, and the other was above average in precipitation, leading to reproductive success in four of five populations. Stochastic simulations of population growth revealed patterns indicative of bet hedging via seed banking, particularly in the annual populations: variance in fitness and fitness in the average environment were lower with seed banking than without, whereas long-run stochastic mean fitness was higher with seed banking than without across a wide range of probabilities of the wet year. This represents a novel, unusually rigorous demonstration of bet hedging from field data.