Information about the world is captured by our separate senses, and must be integrated to yield a unified representation. This raises the issue of which signals should be integrated and which should remain separate, as inappropriate integration will lead to misrepresentation and distortions. One strong cue suggesting that separate signals arise from a single source is coincidence, in space and in time. We measured increment thresholds for discriminating spatial intervals defined by pairs of simultaneously presented targets, one flash and one auditory sound, for various separations.We report a 'dipper function', in which thresholds follow a 'U-shaped' curve, with thresholds initially decreasing with spatial interval, and then increasing for larger separations. The presence of a dip in the audiovisual increment-discrimination function is evidence that the auditory and visual signals both input to a common mechanism encoding spatial separation, and a simple filter model with a sigmoidal transduction function simulated the results well. The function of an audiovisual spatial filter may be to detect coincidence, a fundamental cue guiding whether to integrate or segregate. © Koninklijke Brill NV, Leiden, 2013.