This dissertation explores consumers’ judgments and decisions under uncertainty through the lens of belief distributions. Asking people to consider all possible outcomes and indicate their likelihoods – a practice referred to as “constructing a belief distribution” – has been on the rise in disciplines including marketing, management, psychology, and economics, finding its applications in diverse research topics and professional forecasting. It has been used as an elicitation method to measure people’s beliefs over all possibilities of uncertain outcomes and has been proposed as a light-touch intervention to reduce people’s overconfidence. Each chapter in this dissertation investigates a different aspect of this practice. Chapter 1 examines the effectiveness of belief distributions as an overconfidence intervention. Across different prediction domains, we find that constructing a belief distribution actually increases people’s overconfidence. This is because the process of allocating probabilities to different outcomes is infused with confirmatory reasoning: People tend to allocate probabilities in a way that reinforces, rather than calls into question, their prior beliefs. Chapter 2 turns to an important question in using belief distributions as a measure: Do people construct the same belief distributions regardless of how they are elicited? We find that two functionally similar methods – Distribution Builder and Sliders, both eliciting people’s belief distributions in a graphical way – lead to different results. In particular, the Distribution Builder consistently elicits more accurate responses than the Sliders, in part because those using Sliders tend to start from the first category and end up allocating excessive mass to the starting categories. Chapter 3 applies belief distributions to investigating the communication of uncertainty in time estimates. Across different domains, time durations, and underlying distributions, we find that time estimates presented as ranges increase consumer satisfaction relative to those presented as point estimates. This is in part because range estimates widen people’s anticipated distributions of outcomes and thus expand the interval in which outcomes feel consistent with people’s expectations relative to a counterfactual in which a point estimate has been provided. Together these three investigations shed light on the study of belief distributions, biases in judgments, and consumer decisions under uncertainty.