Pension plan managers are increasingly expected to demonstrate that the risk preferences of plan participants are adequately reflected in the asset allocations of their plans. Yet, it remains unclear whether managers should elicit the subjective risk preferences of individual participants to achieve this goal, or instead rely on objective and more readily available indicators, such as the socio-demographic characteristics of participants. To address this question, we have developed an augmented lottery choice method, tailored to the pension domain, to measure the risk preferences of individuals. Results from 7,894 participants from five different Dutch pension plans show that risk preferences vary substantially, but that only a small fraction of this heterogeneity can be explained by directly observable characteristics. We show that variations in risk preferences, along with individual characteristics and market conditions, generate differences in optimal asset allocations among plans. However, the risk preferences to be elicited for a pension plan depend on the composition of the participant population, as well as on expected market conditions. We provide a framework that pension plan managers can use to gauge the value of elicitation of risk preferences in the population of their funds, so that they can perform their own cost-benefit analysis of the exercise. We illustrate how this framework can be applied by Dutch pension plan managers in deciding on whether to elicit the risk preferences of participants or not.