Decisions under ambiguity depend on the beliefs regarding possible scenarios and the attitude towards ambiguity. This paper exclusively focuses on beliefs, and beliefs are measured independently from attitudes, in contrast to many previous studies. We use laboratory experiments to estimate the subjective belief formation and belief updating process in an ambiguous environment. As a main contribution, we recover the entire be- lief distribution of individual subjects and scrutinize how beliefs are updated in response to new information. For 70% of the subjects, we can reject the objective equality hypothe- sis that one’s initial prior follows a uniform distribution. A further investigation of biases in initial beliefs reveals that 66% of the subjects display neither pessimism nor optimism in initial beliefs. Overall, the unbiased belief hypothesis cannot be rejected. The recovered belief updating rules reveal that the Bayesian updating hypothesis can be rejected for 84% of the subjects. Among them, most subjects under-react to new information compared to what Bayes’ rule implies. Finally, we find that beliefs are heterogeneous and cannot be characterized by a single distribution that fits for all subjects.