The importance of variations in hospital census for planning bed requirements has long been recognized. Traditional models assume Poisson or normal distribution patterns accurately describe such variations in specific hospital units, entire facilities, and groups of hospitals serving the same area. A study of 20 groups of hospitals in five states indicates that these traditional statistical distributions are only occasionally accurate. Different services display different census variation patterns. The standard deviation of the daily census distribution becomes increasingly higher than the square root of the mean estimate as the size of the mean increases. Implications of this finding for regional planning and coordination of hospitals are discussed.