Vertebrates have evolved mechanisms to exploit amplitude modulations in background noise for improving signal detection. However, the mechanisms underlying this masking release are not yet well understood. Here we present evidence for masking release observed in European starlings (Sturnus vulgaris, Aves) that were trained in a Go/NoGo paradigm to report the detection of a short tone (20 ms) in 100% sinusoidally amplitude-modulated noise maskers (400 ms duration). Maskers centred at the tone frequency were composed of one, three, or five spectrally adjacent noise bands each of auditory filter bandwidth. Envelopes of the masking noise bands were either in-phase (i.e. coherent) or successively phase shifted by 90° (i.e. incoherent). A release from masking of up to 28 dB was observed for detection of signals presented in dips of the envelope of coherent maskers compared with those presented in peaks of coherent maskers and in incoherent maskers. For maskers limited to one auditory filter (i.e. limited to the analysis channel tuned to the test signal) this masking release was about 10 dB less than that observed for maskers allowing a comparison across three or five auditory filters. This indicates that both within-channel cues and across-channel cues are important for signal detection. These behavioural data provide the reference for the study of responses of auditory forebrain neurons in the same species reported in a companion paper [Nieder & Klump (2001) Eur. J. Neurosci., 13, 1033–1044].