Observers detected the presence of Gabor signals in fields of two-dimensional broadband Gaussian noise encased by a frame of uniform noise. These images, displayed for 1 sec on a t.v. monitor, were seen through a simple-lens stereoscope. While the left and right eye were presented with the same view of the noise frame, the right-eye Gaussian noise field was right-shifted relative to the left-eye's by 20.3 min arc along the horizontal axis. Useful interocular cues concerning signal presence were not available in one set of trials (control condition), the right-eye signal being displaced by the same amount and direction as that of the noise. In another set of trials (experimental condition), such cues were made available by presenting the Gabor signals in exactly corresponding locations in the centres of the two frames. We tested the predictions of a model in terms of which the usefulness of binocular cues for the "unmasking" of the signal rests upon the characteristics of the summed monocular inputs, specifically the signal-to-noise ratio in the summed pattern. The two-dimensional power spectrum for the summed Gaussian noises shows "notches" at specific horizontal- and vertical-frequency intersections where the power density is at or close to zero. If the spectral power in the summed Gabor signal is concentrated at these locations in the experimental condition, the signal should be unmasked. Accordingly, the spatial frequencies and orientation of three Gabor patterns were chosen in such a way that the power density of the summed signals would fit in notches of the power spectrum of the summed noise, while it would overlap with a noise peak for a fourth Gabor signal. The findings were consistent with the summation hypothesis: binocular masking level differences of up to 18 dB were observed, but only for the three signals "falling" in the two-dimensional noise notches.