These experiments examine the effects of masker level on the magnitude of comodulation masking release (CMR). In experiment 1, threshold was measured for detecting a 2000‐Hz signal in noise bands 100‐ or 3200 Hz wide, centered at the signal frequency. The noise was either amplitude modulated by a low‐pass‐filtered noise, or was unmodulated. At noise spectrum levels of 30 and 50 dB, thresholds were lower in the 3200‐Hz‐wide modulated noise than in the 100‐Hz‐wide modulated noise or the 3200‐Hz‐wide unmodulated noise, indicating a CMR. The magnitude of this CMR decreased at a noise spectrum level of 10 dB, and was very small at a spectrum level of −10 dB. In experiment 2, threshold was measured for a 700‐Hz signal centered in a 20‐Hz wide band of noise (the on‐frequency band, OFB), both in the presence and absence of eight flanking bands (FBs) whose envelopes were either identical with that of the OFB (correlated condition) or were uncorrelated. Thresholds were lower in the correlated than in the uncorrelated condition, indicating a CMR. When the OFB and the FBs were presented to the same ear, the CMR decreased when the spectrum level of all bands was below 30 dB, or when the spectrum level of the FBs was decreased below 40 dB keeping the level of the OFB constant at 40 dB. When the OFB and the FBs were presented to opposite ears, the CMR decreased when the spectrum level of all bands was decreased below 30 dB or when the spectrum level of the FBs was decreased below 40 dB, keeping the level of the OFB fixed at 40 or 60 dB. However, the CMR was almost independent of the spectrum level of the OFB (over the range 10–70 dB) when the spectrum level of the FBs was held constant at 60 dB. The results are interpreted in terms of perceptual grouping mechanisms. Implications for the measurement of CMR in hearing‐impaired subjects are also discussed.