Measurements were made of the masking of pure tones by bands of noise, each band width corresponding to a pitch interval of 250 mels, with band level varied from 40 to 120 db (above 0.0002 microbar). A modified Békésy audiometer permitted recording as a function of frequency, so that masked audiograms could be compared with the noise spectra to ascertain how well the ear followed the frequency characteristics of the noise bands at different levels. It was found that values for the critical band according to Fletcher and to Hawkins and Stevens predict quite precisely the masking for tones whose frequencies lie within the band. More masking than would be predicted is found for frequencies above the band, the discrepancy increasing as the noise level increases. Below the band, the masked threshold is again predicted precisely by the critical‐band hypothesis, but only to a point, beyond which, for high noise levels, a “remote” masking is observed that is constant with frequency for any particular level and is much more than can be predicted from the effective levels (from either system noise or imperfect band rejection) obtaining at those low frequencies where this masking is observed. Whereas masking for tones within the band increases linearly with level, this “remote” masking rises in an accelerated manner, as a function of level, and then shows deceleration at the highest levels. Some speculative interpretation is attempted.