This paper proposes a new method for the recognition of consonants by use of short time frequency spectra and their transition pattern(Time Spectrum Pattern), and shows the effectiveness of the method with the experimental results. The recognition score is90%for3295unvoiced plosives(/p/, /t/, /k/, and/c/)in212kinds of words uttered by 10 male and 10 female speakers. In the experiment, a single reference pattern common to male and female speakers is used for each phoneme. Accurate recognition of consonants is necessary not only for the recognition of continuous speech, but also for the recognition of large number of spoken word.