The binaural masking level difference (BMLD) is a striking and well- documented psychophysical effect which relates to the ability to use the phase of low-frequency sounds to dissociate them from masking noise. When identical tones and noise are presented to both ears, detectability is improved by up to 15 dB simply by inverting the phase of either the tone or noise in one ear. Measurements of BMLDs were made in single delay-sensitive neurones in the inferior colliculus of the guinea pig. These have confirmed and extended an earlier report [D. M. Caird, A. R. Palmer, and A. Rees, Hear. Res. 57, 91-106 (1991)] by demonstrating that when signals are optimized for the frequency, level, and interaural delay sensitivities of each neurone, BMLDs can be measured which are in a direction, and of a magnitude, consistent with appropriate psychophysical observations in human subjects. In addition, BMLDs were found to be consistent with the delay sensitivities of the neurones to the signal and masker, the major determinant of the masked threshold for optimized signals being the activity evoked in the neurone by the masking noise. Within-channel signal-to-noise (S/N) ratios at masked threshold for single neurones varied from +20 to -7 dB, depending on the binaural configuration and the units' delay sensitivities. In single neurones, the size of the BMLD for optimized signals increased with the level of the noise. The BMLD increased by 5 dB over a 40-dB range of noise, consistent with psychophysical observations. This came about because as noise level increased, masked threshold for optimized tones increased more slowly in N(π) noise than in N0 noise. For all binaural comparisons, both positive (π signals more detectable, as in the psychophysics) and negative BMLDs were observed, often in the same neurone, a result entirely consistent with the sensitivity to the interaural delay of the noise and tone signals. For 500- Hz signals in zero and π phase masked by identical noise the majority of BMLDs determined with the PEST procedure was negative, a result which is taken to indicate that increases in spike rate may not be an appropriate cue for masked threshold under these conditions.