Statistical regularities have an influence on perception, as shown by visual search and visual learning^1-3. Here we asked whether objects rendered invisible by binocular rivalry could nonetheless contribute to perception of average size within an array of different sized objects. First, we tested whether observers could estimate the mean size of an array distributed between the two eyes. Two arrays each containing 12 circles were briefly presented on either side of fixation. Observers judged which array had the larger mean size, under three presentation conditions 1) both arrays presented to a single eye, 2) one array presented to one eye and the other array to the other eye, 3) circles from both arrays randomly presented to one eye or the other. Accuracy was equivalently good for all three conditions, implying that mean size is computed at a central, post-binocular site. In the main experiment, observers performed the same task, this time with some circles (either 0, 4, 8, or 12 circles) suppressed from consciousness by the presence of high contrast rival targets in the opposite eye (we verified suppression's effectiveness). In a comparison condition we physically removed circles but left the rival targets, thus mimicking suppression's effect on visibility. Accuracy in the mean discrimination task was higher when some physically present circles were blocked from consciousness than when those circles were physically missing. Thus, statistical regularity can be extracted from a scene containing items that themselves fall outside of conscious awareness.