Reactor noise is a very important source of information on operational characteristics of nuclear power plants. The problem of nuclear power plant noise diagnostics is formulated here as a problem of the pattern recognition theory. The use of the entropy criterion, the difference of the conditional probability density criterion, and the Karhunen-Loeve expansion for feature extraction are discussed. 19 refs.