I consider the elicitation of ambiguous beliefs about an event and show how to identify the interval of relevant probabilities (representing ambiguity perception) for several classes of ambiguity averse preferences. The agent reveals her preference for mixing binarized bets on the uncertain event and its complement under varying betting odds. Under ambiguity aversion, mixing is informative about the interval of beliefs. In particular, the mechanism allows to distinguish ambiguous beliefs from point beliefs, and identifies the belief interval for maxmin preferences. For ambiguity averse smooth second order and variational preferences, the mechanism reveals inner bounds for the belief interval, which are sharp under additional assumptions. In an experimental study, participants perceive almost as much ambiguity for natural events (generated by the stock exchange and by a prisoners dilemma game) as for the Ellsberg Urn, indicating that ambiguity may play a role in real-world decision making.