In probabilistic models, the probabilities are always considered to be known. However, in one experiment we may observe the outcomes but not the corresponding probabilities. What is the connection between the observed outcomes and assigned probabilities, and can the probability be measured exactly?