Adaptation to temporally fluctuating environments can be achieved through direct phenotypic evolution, by phenotypic plasticity (either developmental plasticity or trans-generational plasticity), or by randomizing offspring phenotypes (often called diversifying bet-hedging). Theory has long held that plasticity can evolve when information about the future environment is reliable while bet-hedging can evolve when mixtures of phenotypes have high average fitness (leading to low among generation variance in fitness). To date, no study has studied the evolutionary routes that lead to the evolution of randomized offspring phenotypes on the one hand or deterministic maternal effects on the other. We develop simple, yet general, models of the evolution of maternal effects and are able to directly compare selection for deterministic and randomizing maternal effects and can also incorporate the notion of differential maternal costs of producing offspring with alternative phenotypes. We find that only a small set of parameters allow bet hedging type strategies to outcompete deterministic maternal effects. Not only must there be little or no informative cues available, but also the frequency with which different environments are present must fall within a narrow range. By contrast, when we consider the joint evolution of the maternal strategy and the set of offspring phenotypes we find that deterministic maternal effects can always invade the ancestral state (lacking any form of maternal effect). The long-term ESS may, however, involve some form of offspring randomization, but only if the phenotypes evolve extreme differences in environment-specific fitness. Overall we conclude that deterministic maternal effects are much more likely to evolve than offspring randomization, and offspring randomization will only be maintained if it results in extreme differences in environment-specific fitness.