The given article deals with the genetic algorithm of creating specialized filters for undimensional biomedical signals analysis. We offer several ways of action for each stage of the given algorithm's performance. We have shown the possibility of comparing both average modules of values of the difference between the coefficients of the convolution with such filters of two sets of EEG signals that correspond to the two explored states, and modules with a maximum difference of values of convolution coefficients. We offer two approaches of averaging the filters over the course of their "crossing": in the time and frequency domains. The advantages and disadvantages of each of these approaches are discussed. To carry out the "mutation", we apply noises: additive blue, flicker and/or white noise, and for "crossover" — the exchange of time sequences for the filters. There is demonstrated a successful implementation of the given algorithm for defining two states of EEG, received during the experiments with systems of human-machine interaction. We demonstrated that already by the 20th generation of filters the difference of coefficients of signals convolution with bred filters significantly increases compared to first generations. The offered approach lets drastically increase the research opportunities white detecting different functional states in humans and animals.