Aims: Optimal decision making in a volatile environment relies on the brain's representation of uncertainty. Expected utility provides a normative theory for explaining decision outcomes under uncertainty, though its limitations have prompted the exploration of other preference measures. Specifically, we do not yet fully understand which preference measure the brain uses for probabilistic decision making, and how dopaminergic system might serve to implement this measure. Here we suggest a new approach to find this preference measure and show how it can be computed within a population of dopaminergic neurons.Methods: We assume that the reward distribution is represented by the expected values of the encoding functions of dopaminergic neurons. We show that this representation, referred to as distributed distributional coding (DDC), can be recursively updated based on recently sampled reward outcomes. We propose a DDC network and a learning rule to decipher the utility functions and coherent risk/preference measures of the brain. Results: Our method determines any utility function that the brain may employ for decision making, and demonstrates that a weighted sum of the responses of dopaminergic neurons can approximate the expected utility. We also show that the DDC network can model coherent risk measures, such as conditional value at risk (CVaR). Lastly, we show how a population of dopaminergic neurons can compute these coherent measures.Conclusions: Our framework provides a novel perspective on the preference measures of the brain and shows how necessary probabilistic measures can be implemented in the dopaminergic system.