We developed a method for estimating the mean and standard deviation of ratios of normal vertebral heights from a sample that includes people with and without vertebral fractures. This method assumes that the measurements in normal vertebrae have a Gaussian distribution and that, for any vertebral level, the prevalence of abnormal measurements is less than 10%. Under these assumptions, normal values for nonfractured vertebrae can be estimated from several statistical properties of Gaussian distributions. We applied these methods to the lateral spinal radiographs of 2992 women aged 65-70 years who were recruited from population-based listings. The estimated means and standard deviations for ratios of dimensions in nonfractured vertebrae were very similar to those based on studies of premenopausal women. Our method may be useful for defining normal values from large populations that include normal and abnormal women, does not require x-rays of normal premenopausal women, avoids the potential biases of defining normality based on qualitative judgment, and can be applied to other types of physical and biochemical measurements.