The purpose of this study was to investigate the influence of within subject and within tooth variability of attachment-level measurements on hypothesis testing. Full-mouth attachment-level measurements were obtained at 4 sites per tooth in early onset periodontitis subjects (both localized N = 89 and generalized N = 139) and adult periodontitis subjects (N = 309). Variance component models utilizing iterative generalized least squares were employed to estimate the % of variance distributed at the level of the site, tooth and subject in the 3 subject populations. The data indicate that a significant % of the variance is distributed within the tooth and the subject. Ignoring the variance attributable to teeth or subjects can result in inappropriately low type I error rates for hypothesis testing. Thus, both the subjects and the tooth must be considered in the modeling of attachment level measurements. Also in studies in which a limited number of samples are taken, data analysis would be simplified if these samples were taken from different teeth.