The paper addresses a non-traditional problem of pattern recognition, when information about pattern is represented in the form of a random signal taken from the output of a corresponding physical sensor. It is supposed that there exist two types of signal to recognize, namely, specified in the statistical sense signals and totally unknown signals. Such the conditions are called conditions of increased a priory uncertainty. Developing a technique to recognize specified signals in conditions of increased a priory uncertainty is the objective of this paper. Methods for selection and recognition of a statistically defined random signal are proposed for the cases when signal description is done by various probabilistic models. Additional consideration is given to peculiarities of employing these methods for solving applied problems of pattern recognition in radar, medical diagnostics and speaker identification.