The best strategy for an organism to deal with unpredictable environmental conditions is a stochastic one, but it is not easy to distinguish it from nonadaptive randomness in phenotype production, and its convincing demonstrations are lacking. Here we describe a new method for detection of adaptive stochastic polyphenism and apply it to the following problem. In fall, each female of the bird cherry–oat aphid, Rhopalosiphum padi, faces a decision either to produce sexuals, which mate and lay cold‐tolerant eggs, or to continue production of cold‐sensitive parthenogenetic females, which potentially yields a higher population growth rate but is risky because a cold winter can kill all of her descendants. Using a simulation model, we show that global investment in sexual reproduction should be proportional to winter severity and that variance in the peak date of production of sexual individuals should depend on climate predictability. Both predictions are validated against standardized trap data on aphid flight accompanied by meteorological data, and the predictions support adaptive phenotypic plasticity