The present study examined the flexibility of the auditory system to integrate energy spread over a band of frequencies. Minimum spectral integration can be measured when a single tone is masked by various bandwidths of noise (a “critical‐band” experiment); maximum spectral integration can be examined when a signal noise, of various bandwidths, is masked by wideband noise. The present study compares these two measures directly by employing the same stimuli and experimental procedures. Digitally synthesized noise, with essentially rectangular spectra (3000‐dB/octave skirts), was used either as masker or signal. In the first case, when the masking noise bandwidth was varied, just‐detectable signal level for a 1000‐Hz tone increased at 1.7 dB/octave untilWm= 70–110 Hz and then remained essentially constant. In the second case, when signal bandwidth (centered at 1000 Hz) was varied, just‐detectable signal level was essentially constant untilWs= 70–110 Hz and then increased at 1.7 dB/octave. The coincidence of the two breakpoints suggests a simple critical‐band estimation procedure derived from two thresholds: a tone in wideband noise and a wideband signal in a wider‐bandwidth noise. Based on the predictions for the energy‐detector model, a 1.5‐dB/octave slope implies an efficient spectral integration over the signal bandwidth. Combined with Green's similar slope, the present results indicate that the maximum limit of spectral integration may be at least 3 kHz wide. Additional results suggest abrupt spectral edges in masking noise without a wideband background slightly interferes with a simultaneous signal; the effect is similar to nonsimultaneous suppression/inhibition effects, although the cause may not be the same. [Work supported by NIH.]