Results from large-scale surveys indicate that a key reason underlying COVID-19 vaccine refusal are concerns regarding the side effects and the benefits of the vaccines. However, the cognitive mechanisms related to vaccine refusal are unknown. Since the choice to get vaccinated is a risky one (vaccination outcomes are probabilistic), it can be understood through the lenses of cognitive theories of risky decision-making. Here, we examine to what extent vaccine refusal is related to probability neglect – a tendency to ignore probabilities of highly aversive outcomes, such as extreme side effects (e.g., blood clots). In an online study, we contrasted the behavior of three participant groups with different COVID-19 vaccination attitudes: Anti-vaccine (n = 365), Neutral (n = 373), or Pro-vaccine (n = 462). Each participant made a series of hypothetical accept-reject choices about eight existing COVID-19 vaccines. For each vaccine, we presented real data on possible side effects and benefits using the process-tracing methodology Mouselab, which allowed us to monitor predecisional information search. First, we observed that among participants in the Neutral and Pro-vaccine groups and for vaccines with rare and extreme side effects, vaccine rejection was higher than in trials without probability neglect. Second, using computational modeling, we show that the prevalence of vaccine refusal in the Anti-vaccine group is linked to extreme, negative emotional evaluations of the vaccines’ side effects and overall neglect of probabilities. These results suggest a need for individualized interventions aiming to increase vaccination uptake.