The conducted research studied a feature extraction method in a musical instrument tone recognition system. The purpose of this study was to obtain a number of feature extraction coefficients that are smaller than those obtained in previous studies. The studied feature extraction was a DCT (Discrete Cosine Transform)-based segment averaging and Gaussian windowing. The testing of the musical instrument's tone recognition system was carried out using pianica, tenor recorder, and bellyra musical instruments, each of which represented many, several, and one significant local peaks in the transform domain. The test results showed that the optimal number of feature extraction coefficient was 8 coefficients, which could give a recognition rate of up to 100%. The test results were achieved using a Gaussian window with a sigma value of 2-6, and a 128 points DCT.