The aim of this study was to assess and quantify the random effects resulting from clustering in the following individual-level periodontal outcomes: presence of clinical attachment loss of > or = 1 mm (CAL1), presence of clinical attachment loss of > or = 3 mm (CAL3), and presence of necrotizing ulcerative gingivitis (NUG); or in the following class-level periodontal outcomes: number of students with CAL1, number of students with CAL3, and number of students with NUG. Mixed-effects logistic regression analysis was used to model these outcomes among 9,162 adolescents in 310 classes in 98 schools spread over 20 communes in the Province of Santiago, Chile, who had been examined for clinical attachment level and NUG, and had completed questionnaires on oral health-related behaviors. The results of all six analyses demonstrated statistically significant random effects, which in all analyses were particularly related to the schools, whereas the class effects were smaller and the commune random effects were almost negligible. The random effects were quantified using the median odds ratio (MOR), and the class-level MOR ranged between 1.05 and 1.51, whereas the school-level MOR values ranged from 2.07 to 2.39. The results of the study demonstrate the potential of the application of multilevel modeling to periodontal epidemiologic data, over and beyond the conventional use of the technique to account for the intrinsic sites-teeth-subject hierarchy in periodontal data.