Visual figure–ground segregation is achieved by exploiting differences in features such as luminance, colour, motion or presentation time between a figure and its surround. Here we determine the shortest delay times required for figure–ground segregation based on purely temporal features. Previous studies usually employed stimulus onset asynchronies between figure- and ground-containing possible artefacts based on apparent motion cues or on luminance differences. Our stimuli systematically avoid these artefacts by constantly showing 20 × 20 ‘colons’ that flip by 90° around their midpoints at constant time intervals. Colons constituting the background flip in-phase whereas those constituting the target flip with a phase delay. We tested the impact of frequency modulation and phase reduction on target detection. Younger subjects performed well above chance even at temporal delays as short as 13 ms, whilst older subjects required up to three times longer delays in some conditions. Figure–ground segregation can rely on purely temporal delays down to around 10 ms even in the absence of luminance and motion artefacts, indicating a temporal precision of cortical information processing almost an order of magnitude lower than the one required for some models of feature binding in the visual cortex [e.g. Singer, W. (1999), Curr. Opin. Neurobiol., 9, 189–194]. Hence, in our experiment, observers are unable to use temporal stimulus features with the precision required for these models.