The purpose of the present investigation was to evaluate methods to detect periods of destructive periodontal disease activity in individual sites using pairs of repeated attachment level measurements. Attachment level measurements were made at 6 sites on every tooth in 22 individuals with radiographic evidence of periodontal destruction, and were repeated within 7 days. A total of 3414 sites were monitored at 2-month intervals for approximately 1 year. 3 analytical procedures were used to test for significant changes in attachment level. For regression analysis, a linear least squares fit function of time in days vs attachment level was computed for each site and the slope tested for difference from 0. Running medians of 3 were used to smooth attachment level measurements and changes greater than 2 mm in the smoothed curves were considered significant. By the tolerance method, differences between pairs of attachment level measurements were used to compare the mean change and the site specific variability of that change. The proportion of specific agreement (Ps) for breaking down sites was highest between the tolerance and running median methods (Ps = 0.63). Overall agreement (kappa), which included sites which showed "loss", "gain", and no change was 0.56. By regression analysis (P less than 0.01), 175 sites were identified as having significant attachment loss and 79 sites were identified as improving. By running medians these figures were 90 and 50, and by tolerance 94 and 40, respectively. Each of the 3 methods had certain advantages. Regression analysis was particularly sensitive to gradual changes in slope whereas the running median method detected abrupt changes in attachment level. The tolerance method was well suited to detecting changes over a short period of time. The tolerance and running median methods detected more breaking down sites on the molars and lower incisors and on interproximal surfaces; whereas regression analysis did not show these differences.