Introduction. The issues of features informativity evaluation in biomedical signals portraits according to formal comparison of their probability values, obtained during training of supervised learning recognition systems are being considered. The purpose of this work is to increase the effectiveness of the biomedical signals recognition in diagnostic systems with supervised learning, by choosing rational structure of portraits based on the nature of their elements influence on the quality of pattern recognition.Details. Informativity values are used for feature selection in the formation of truncated portraits. Subsequent exclusion of less informative features from consideration relies in this paper in the basis of disclosure and implementation of reserves to increase the probability of correct recognition in diagnostic systems. Content and efficiency of the proposed signal processing technology is illustrated by the test case in the application to the common task of recognition of QRS-complexes types, useful for recognition system training with its data selection.Conclusions. The use of certain components of biomedical signals portraits, studied in the diagnosis of patients can have a positive impact on the quality of pattern recognition, while others create redundancy of portraits or reducing effectiveness of solving this task.