The threshold for detecting a narrow-band noise signal in the presence of one or more masking noise bands is higher when the signal and masker bands have the same envelope (i.e., are comodulated) than when they have independent envelopes. This is called a comodulation detection difference (CDD). CDD might be caused by perceptual grouping of the signal and masker bands when they are comodulated. This hypothesis leads to the prediction that some masking should occur for comodulated bands, even when they are widely separated in frequency. An alternative hypothesis is that CDD occurs because, when the signal and masker bands are independent, the signal band can be detected in the dips of the masker envelope. This leads to the prediction that CDD should only occur when the masker produces significant excitation at the signal place. Experiment 1 tested these predictions in a paradigm similar to two-tone masking. The signal was a 20-Hz-wide noise centered at 1500 Hz, and the masker consisted of two bands of noise on either side of the signal frequency, whose envelopes were either comodulated (condition C) or uncorrelated (condition U) with the envelope of the signal band. In a third condition (S), the signal band was replaced by a sinusoid. The frequency separation between the signal and masker bands, delta f, was varied from 100 to 1400 Hz. Thresholds were very similar for conditions U and S; thresholds declined progressively as delta f increased beyond 200 Hz, and reached the absolute threshold for delta f = 1400 Hz. For values of delta f from 200 to 1000 Hz, thresholds were higher for condition C than for conditions U or S (i.e., a CDD occurred), but for delta f = 1400 Hz thresholds for condition C also reached absolute threshold. In experiment 2, delta f was fixed at 600 Hz and conditions were included where only the upper or the lower masker band was correlated with the signal band. Also, the overall level of the masker was systematically varied. The results indicate that the magnitude of CDD is determined by the comodulation of the signal band with the masker band producing the most masking. Overall, the results support an explanation based on the spread of excitation and dip listening, rather than an explanation based on perceptual grouping.