In order to clarify the characteristics of frequency normalization on voiceless consonants in which the frequency axes were compressed or expanded, perceptual experiments were carried out for voiceless fricatives and voiceless stops. For voiceless fricatives, a series of fricative noises that varied from the center frequency appropriate for /// to one appropriate for /s/ were synthesized as stimuli. These stimuli were subjected to a listening test to determine the phonetic categorical boundary between //a/ and /sa/ on the continuum of the noise frequency for various frequency-compression or expansion rates. For voiceless stops, noise part was simulated by either a single-pole noise or a multiple-pole noise. A series of noises that varied from /k/ to /t/ were synthesized and subjected to a identification test between /ka/ and /ta/. The results showed that the categorical boundaries between //a/ and /sa/ were invariant, but the categorical boundaries between /ka/ and /ta/ moved in relation to the frequency-compression or expansion rate for both noise conditions. These results implied that the frequency normalization was made on voiceless stops but not on voiceless fricatives.