Our recent study in Gansu Province, China reported an increasing risk of lung cancer with increasing residential radon concentration that was consistent with previous pooled analyses and with meta-analyses of other residential studies (Wang et al., Am. J. Epidemiol. 155, 554-564, 2002). Dosimetry used current radon measurements (1-year track-etch detectors) in homes to characterize concentrations for the previous 30 years, resulting in uncertainties in exposure and possibly reduced estimates of disease risk. We conducted a 3-year substudy in 55 houses to model the temporal and spatial variability in radon levels and to adjust estimates of radon risk. Temporal variation represented the single largest source of uncertainty, suggesting the usefulness of multi-year measurements to assess this variation; however, substantial residual variation remained unexplained. The uncertainty adjustment increased estimates of the excess odds ratio by 50-100%, suggesting that residential radon studies using similar dosimetry may also underestimate radon effects. These results have important implications for risk assessment.