Geographic modeling is increasingly being used to estimate long-term environmental exposures in epidemiologic studies of chronic disease outcomes. However, without validation against measured environmental concentrations, personal exposure levels, or biologic doses, these models cannot be assumed a priori to be accurate. This article discusses three examples of epidemiologic associations involving exposures estimated using geographic modeling, and identifies important issues that affect geographically modeled exposure assessment in these areas. In air pollution epidemiology, geographic models of fine particulate matter levels have frequently been validated against measured environmental levels, but comparisons between ambient and personal exposure levels have shown only moderate correlations. Estimating exposure to magnetic fields by using geographically modeled distances is problematic because the error is larger at short distances, where field levels can vary substantially. Geographic models of environmental exposure to pesticides, including paraquat, have seldom been validated against environmental or personal levels, and validation studies have yielded inconsistent and typically modest results. In general, the exposure misclassification resulting from geographic models of environmental exposures can be differential and can result in bias away from the null even if non-differential. Therefore, geographic exposure models must be rigorously constructed and validated if they are to be relied upon to produce credible scientific results to inform epidemiologic research. To our knowledge, such models have not yet successfully predicted an association between an environmental exposure and a chronic disease outcome that has eventually been established as causal, and may not be capable of doing so in the absence of thorough validation.