The influence of several clinical and microbiological variables on the site-specific risk of attachment loss was studied in Navajo Indian adolescents aged 14-19. Diagnoses were made at mesio-buccal sites of the four first permanent molars. Case-control analytical methods were used, with A. actinomycetemcomitans, B. gingivalis, and B. intermedius considered the "risk" variables, and with calculus, gingival bleeding, age, and gender treated as possible confounders. The presence of B. intermedius significantly increased the likelihood that attachment loss would be diagnosed at a site (odds ratio = 2.86). However, this association was confounded by calculus and gingival bleeding; when either or both were present, the effect of B. intermedius was markedly weaker. Step-wise multiple logistic regression analyses showed that, of the variables considered, the combination of calculus, gingival bleeding, and B. intermedius gave the most parsimonious explanation of the presence of attachment loss. The chance that attachment loss would be diagnosed was increased five times when calculus was present, 16.5 times in the presence of both calculus and gingival bleeding, and 37 times when these variables plus B. intermedius were observed at a particular site.