We assessed the reproducibility over a 6-year period of 16 trace elements measured in toenails by comparing levels in paired specimens collected in 1982-1983 and 1988 from 127 women in the United States. The Spearman correlation coefficients for the reproducibility of toenail levels of selenium and arsenic (both known to reflect intake of these elements) were 0.48 and 0.54. Correlations for other elements ranged from 0.26 (copper) to 0.58 (zinc). In utilizing biomarkers to assess exposure in epidemiological studies of cancer and other chronic disease, random within-person variability in exposure leads to attenuation of measures of association between exposure and disease. We demonstrate the effect of such variability on odds ratios from a hypothetical case-control study. For a true odds ratio of 3.0 (for a comparison of the highest quintile versus the remaining 4 quintiles of exposure) the odds ratios which would be observed in the presence of the degree of within-person variability demonstrated in this study were 2.15 for toenail arsenic and 1.67 for toenail copper levels. Toenail concentrations of certain trace elements are useful biomarkers of exposure in which a single sample is assumed to represent long-term exposure. However, substantial attenuation in measures of association may occur.