Two experiments investigated the shape of the binaural temporal window using a detection task. In experiment 1, a 10-ms tone burst was presented binaurally out-of-phase during a burst of white noise, which changed from being interaurally uncorrelated, to correlated, and back to uncorrelated. The tone occurred during the correlated portion of the noise in one interval of each 2I-FC trial. Detection thresholds were recorded using a 2-down/1-up adaptive procedure. Thresholds were measured for different durations of correlated noise (0–960 ms), frequencies of tone burst (125, 250, 500, and 1000 Hz) and levels of noise [20, 30, 40, and 50 dB(SPL)/Hz]. Window shapes based on nine candidate functions were fitted to the data using the assumption that the binaural masking release was related to the overall interaural correlation of noise admitted by the window. Fitted windows included both a forward and a backward lobe. Gaussian functions tended to give closer fits than exponential and rounded-exponential functions, and simple functions gave more parsimonious fits that those which included dynamic-range-limiting terms. Using simple Gaussian fits, the shape of the window was largely independent of frequency and level, and the windows for individual listeners had equivalent rectangular durations ranging from 55 to 188 ms. The asymmetry was variable, although forward lobes were generally shorter than backward lobes. Experiment 2 ruled out the possibility that the forward lobe might be an artefact caused by distraction of the listener, when the interaural phase change in the noise closely followed the signal. In this experiment, the out-of-phase tone was presented during a burst of partially correlated noise which changed, after a variable interval, to a fully correlated noise. Thresholds for detecting the tone rose (i.e., performance worsened) as the interval was increased. Distraction would have produced the opposite effect.