We investigated to use mixed features obtained with K-L (Karhunen-Loeve) transform and Fisher's linear discriminant (discriminant analysis) for the recognition of plosives. In the experiment of speaker-independent plosive recognition using the mixed features obtained from LPC cepstrum and PARCOR coefficients, the recognition rate improved in the rate of 3% as compared with that using the features from LPC cepstrum coefficients. Also, in the experiment using the mixed features obtained from spectrum and PARCOR coefficients, the recognition rate improved in the rate of 3% as compared with that using the features from spectrum.