In epidemiological studies, investigators have routinely used ambient air concentrations, measured by air-quality monitoring networks, to assess exposure of subjects. When there is great spatial variability of ambient air concentrations or when there are specific indoor exposures, this approach may yield substantial exposure misclassification and distort the associations between exposure and the health endpoints of interest. In 3 French metropolitan areas, the cross-sectional relationships between 48 hr of nitrogen dioxide personal exposure of 73 children and the corresponding 48-hr background ambient air concentrations were analyzed. The crude correlation between ambient air concentrations and personal exposures was poor in all cities (r2 = .009 for Grenoble, r2 = .04 for Toulouse, and r2 = .02 for Paris). These correlations were improved when the authors took into account other ambient air or indoor air sources of nitrogen dioxide emissions (the corresponding multiple linear regression, r2, increased to .43 in Grenoble, .50 in Toulouse, and .37 in Paris). The main variables that explained personal exposures were an index of traffic intensity and proximity and use of a gas cooker at home. The results of this study confirm that ambient air-monitoring site measurements are poor predictors of personal exposure. Investigators should carefully characterize the proximity of roads occupied by dense traffic to the home/school as well as indoor sources of nitric oxide emissions; both of these careful characterizations will assist researchers in the prediction of personal exposure in epidemiological studies.