Speech band was divided into 8 channels by filters of which midfrequencies were 350, 523, 783, 1174, 1760, 2637, 3591 and 5916 c/s. From the statistical results of analysis obtained from vowels of 100 male students, Japanese vowels can be expected to be discriminated with considerably high accuracy with only first five filters. An experimental vowel discriminator was constructed and examined on 20 men's vowels, whose age ranging from 16 to 60 years. Good results were obtained as expected.