Causal mechanisms linking exposures to disease risks operate in a biological hierarchy running from low-level (molecular and gene-level, cell-level or intracellular events) to intermediate (target cell population and organ- or tissue-level damage) to high-level (age-specific hazard functions for individuals and populations) descriptions of responses to exposures. For many diseases, including lung cancer and chronic obstructive pulmonary disease (COPD), mechanisms of disease induction and progression are not yet fully understood, yet some information is available at each level. A way is needed to synthesize information across levels and to use constraints from observations at all levels to characterize dose-response relations that are consistent with the data. This paper suggests such an approach, which we call multilevel disease modeling (MDM). We illustrate it using lung cancer and COPD risks from smoking as examples. MDM starts with a traditional systems biology markup language (SBML) simulation model of tissue-level events such as genetically altered patch and field formation for lung cancer; or alveolar wall damage and apoptosis for COPD. These systems dynamics models represent cell populations and patches of tissues as making transitions among successively deteriorating stages (corresponding to accumulation of specific markers of damage at the cellular and sub-cellular levels). This basic description of disease progression is augmented with submodels that derive transition rates among stages as outputs of lower-level events (e.g., first-passage times through networks of gene-level changes). Uncertainty is constrained by higher-level observations (e.g., exposure-dependent transition rates among stages must be consistent with epidemiological data on disease progression rates.) We show how to use these multilevel constraints to develop small sets of model specifications that are consistent with all available data, and to predict exposure-dependent risks.